fmaury/seirdy.one
1
0
Fork 0
mirror of https://git.sr.ht/~seirdy/seirdy.one synced 2024-12-24 01:42:10 +00:00
Code Issues Projects Releases Wiki Activity

Move Petal to graveyard, update other engines.

Browse source
This commit is contained in:
Rohan Kumar 2023-08-15 21:54:37 -07:00
parent 4ee4730cb8
commit 32cd723847
No known key found for this signature in database
GPG key ID: 1E892DB2A5F84479
2 changed files with 20 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
content/posts/search-engines-with-own-indexes.gmi
View file

@ -70,10 +70,6 @@ These are large engines that pass all my standard tests and more.
4. Mojeek: Seems privacy-oriented with a large index containing billions of pages. Quality isn’t at Google/Bing/Yandex’s level, but it’s not bad either. If I had to use Mojeek as my default general search engine, I’d live. Partially powers eTools.ch. At this moment, I think that Mojeek is the best alternative to GBY for general web search.
5. Petal search: A search engine by Huawei that recently switched from searching for Android apps to general search. Despite its surprisingly good results, I wouldn't recommend it due to privacy concerns: its privacy policy describes advanced fingerprinting metrics, and it doesn't work without JavaScript. Requires an account to submit sites. I discovered this via my access logs. Be aware that in some jurisdictions, it doesn't use its own index: in Russia and some EU regions it uses Yandex and Qwant, respectively.
=> https://petalsearch.com/ petalsearch.com
Google, Bing, and Yandex support structured data such as microformats1, microdata, RDFa, Open Graph markup, and JSON-LD. Yandex's support for microformats1 is limited; for instance, it can parse h-card metadata for organizations but not people. Open Graph and Schema.org are the only supported vocabularies I'm aware of. Mojeek is evaluating structured data; it's interested in Open Graph and Schema.org vocabularies.
### Smaller indexes or less relevant results
@ -177,7 +173,8 @@ Slzii.com: A new web portal with a search engine. Has a tiny index dominated by
Engines in this category fall back to GBY when their own indexes don't have enough results. As their own indexes grow, some claim that this should happen less often.
* Brave Search: Many tests (including all the tests I listed in the "Methodology" section) resulted results identical to Google, revealed by a side-by-side comparison with Google, Startpage, and a Searx instance with only Google enabled. Brave claims that this is due to how Cliqz (the discontinued engine acquired by Brave) used query logs to build its page models and was optimized to match Google.¹⁰ The index is independent, but optimizing against Google resulted in too much similarity for the real benefit of an independent index to show. Furthermore, many queries have Bing results mixed in; users can click an "info" button to see the percentage of results that came from its own index. The independent percentage is typically quite high (often close to 100%) but can drop for advanced queries.
* Brave Search: Many tests (including all the tests I listed in the "Methodology" section) resulted results identical to Google, revealed by a side-by-side comparison with Google, Startpage, and a Searx instance with only Google enabled. Brave claims that this is due to how Cliqz (the discontinued engine acquired by Brave) used query logs to build its page models and was optimized to match Google.¹⁰ The index is independent, but optimizing against Google resulted in too much similarity for the real benefit of an independent index to show. Furthermore, many queries have Bing results mixed in; users can click an "info" button to see the percentage of results that came from its own index. The independent percentage is typically quite high (often close to 100%) but can drop for advanced queries. (update: Brave's Bing contract appears to have expired as of April 2023).
I can't in good conscience recommend using Brave Search, as the company runs cryptocurrency, has held payments to creators without disclosing that creators couldn't receive rewards, has made dangerously misleading claims about fingerprinting resistance (will update with a link to my thoughts on the matter), is run by a CEO who spent thousands of dollars opposing gay marriage, and has rewritten typed URLs with affiliate links.
=> https://search.brave.com/ Brave Search
@ -202,11 +199,12 @@ These indexing search engines don’t have a Google-like “ask me anything” e
### Small/non-commercial Web
* Marginalia Search: A recent addition similar to Wiby, and *my favorite entry on this page*. It has its own crawler but is strongly biased towards non-commercial, personal, and/or minimal sites. It's a great response to the increasingly SEO-spam-filled SERPs of GBY. Partially powers Teclis, which in turn partially powers Kagi. Update 2022-05-27: Marginalia.nu is now open source
* Teclis: A project by the creator of Kagi search. Uses its own crawler that measures content blocked by uBlock Origin, and extracts content with the open-source article scrapers Trafilatura and Readability.js. This is quite an interesting approach: tracking blocked elements discourages tracking and advertising; using Trafilatura and Readability.js encourages the use of semantic HTML and Semantic Web standards such as microformats, microdata, and RDFa. It claims to also use some results from Marginalia. This has been down for maintenance for a while; I may end up having to move it to the Graveyard section.
* Teclis: A project by the creator of Kagi search. Uses its own crawler that measures content blocked by uBlock Origin, and extracts content with the open-source article scrapers Trafilatura and Readability.js. This is quite an interesting approach: tracking blocked elements discourages tracking and advertising; using Trafilatura and Readability.js encourages the use of semantic HTML and Semantic Web standards such as microformats, microdata, and RDFa. It claims to also use some results from Marginalia. The Web interface has been shut down, but its standalone API is still available for Kagi customers.
=> https://search.marginalia.nu/ search.marginalia.nu
=> https://memex.marginalia.nu/log/58-marginalia-open-source.gmi Announcement: marginalia.nu goes open source
=> http://teclis.com/ Teclis
=> https://kagifeedback.org/d/1838-teclis-is-broken Teclis free version shutdown notice
### Site finders
@ -337,6 +335,9 @@ Time for my first Gemini-exclusive content! A Gemini page about search engines w
These engines were originally included in the article, but have since been discontinued.
* Petal search: A search engine by Huawei that recently switched from searching for Android apps to general search. Despite its surprisingly good results, I wouldn't recommend it due to privacy concerns: its privacy policy describes advanced fingerprinting metrics, and it doesn't work without JavaScript. Requires an account to submit sites. I discovered this via my access logs. Be aware that in some jurisdictions, it doesn't use its own index: in Russia and some EU regions it uses Yandex and Qwant, respectively. Inaccessible as of June 2023.
=> https://web.archive.org/web/20230118225122/https://www.petalsearch.com/ petalsearch.com
* Neeva: formerly in the "semi-independent" section. Combined Bing results with results from its own index. Bing normally isn't okay with this, but Neeva was one of few exceptions. Results were mostly identical to Bing, but original links not found by Bing frequently popped up. Long-tail and esoteric queries were less likely to feature original results. Required signing up with an email address or OAuth to use, and offered a paid tier with additional benefits.
=> https://web.archive.org/web/20230528051432/https://neeva.com/blog/may-announcement Neeva shutdown announcement

22
content/posts/search-engines-with-own-indexes.md
View file

@ -101,9 +101,6 @@ Yandex
[Mojeek](https://www.mojeek.com/)
: Seems privacy-oriented with a large index containing billions of pages. Quality isn't at GBY's level, but it’s not bad either. If I had to use Mojeek as my default general search engine, I'd live. Partially powers [eTools.ch](https://www.etools.ch/). At this moment, _I think that Mojeek is the best alternative to GBY_ for general search.
[Petal Search](https://petalsearch.com/)
: A search engine by Huawei that recently switched from searching for Android apps to general search in order to reduce dependence on Western search providers. Despite its surprisingly good results, I wouldn't recommend it due to privacy concerns: its privacy policy describes advanced fingerprinting metrics, and it doesn't work without JavaScript. Requires an account to submit sites. I discovered this via my access logs. Be aware that in some jurisdictions, it doesn't use its own index: in Russia and some EU regions it uses Yandex and Qwant, respectively.
Google, Bing, and Yandex support structured data such as microformats1, microdata, RDFa, Open Graph markup, and JSON-LD. Yandex's support for microformats1 is limited; for instance, it can parse `h-card` metadata for organizations but not people. Open Graph and Schema.org are the only supported vocabularies I'm aware of. Mojeek is evaluating structured data; it's interested in Open Graph and Schema.org vocabularies.
### Smaller indexes or less relevant results
@ -207,7 +204,9 @@ Engines in this category fall back to GBY when their own indexes don't have enou
[Brave Search](https://search.brave.com/)
: Many tests (including all the tests I listed in the "Methodology" section) resulted results identical to Google, revealed by a side-by-side comparison with Google, Startpage, and a Searx instance with only Google enabled. Brave claims that this is due to how Cliqz (the discontinued engine acquired by Brave) used query logs to build its page models and was optimized to match Google.[^9] The index is independent, but optimizing against Google resulted in too much similarity for the real benefit of an independent index to show. Furthermore, many queries have Bing results mixed in; users can click an "info" button to see the percentage of results that came from its own index. The independent percentage is typically quite high (often close to 100% independent) but can drop for advanced queries.
: Many tests (including all the tests I listed in the "Methodology" section) resulted results identical to Google, revealed by a side-by-side comparison with Google, Startpage, and a Searx instance with only Google enabled. Brave claims that this is due to how Cliqz (the discontinued engine acquired by Brave) used query logs to build its page models and was optimized to match Google.[^9] The index is independent, but optimizing against Google resulted in too much similarity for the real benefit of an independent index to show. Furthermore, many queries have Bing results mixed in; users can click an "info" button to see the percentage of results that came from its own index. The independent percentage is typically quite high (often close to 100% independent) but can drop for advanced queries. <ins cite="https://brave.com/search-independence/" datetime="2023-08-15T20:39:00-07:00">Update 2023-08-15: Brave's Bing contract appears to have expired as of April 2023.</ins>
I can't in good conscience recommend using Brave Search, as the company runs cryptocurrency, has [held payments to creators without disclosing that creators couldn't receive rewards](https://brave.com/rewards-update/), has made dangerously misleading claims about fingerprinting resistance,[^10] is run by a CEO who [spent thousands of dollars opposing gay marriage](https://arstechnica.com/information-technology/2014/03/new-mozilla-ceo-issues-statement-expresses-sorrow-for-causing-pain/), and [has rewritten typed URLs with affiliate links](https://www.pcmag.com/news/brave-browser-caught-redirecting-users-through-affiliate-links).
[Plumb](https://plumb.one/)
: Almost all queries return no results; when this happens, it falls back to Google. It's fairly transparent about the fallback process, but I'm concerned about _how_ it does this: it loads Google's Custom Search scripts from `cse.google.com` onto the page to do a client-side Google search. This can be mitigated by using a browser addon to block `cse.google.com` from loading any scripts. Plumb claims that this is a temporary measure while its index grows, and they're planning on getting rid of this. Allows submitting URLs, but requires solving an hCaptcha. This engine is very new; hopefully as it improves, it could graduate from this section. Its Chief Product Officer [previously founded](https://archive.is/oVAre) the Gibiru search engine which shares the same affiliates and (for now) the same index; the indexes will diverge with time.
@ -229,7 +228,7 @@ These indexing search engines don’t have a Google-like “ask me anything” e
: _My favorite entry on this page_. It has its own crawler but is strongly biased towards non-commercial, personal, and/or minimal sites. It's a great response to the increasingly SEO-spam-filled SERPs of GBY. Partially powers Teclis, which in turn partially powers Kagi. <ins cite="https://memex.marginalia.nu/log/58-marginalia-open-source.gmi" datetime="2022-05-28T14:09:00-07:00">Update 2022-05-28: [Marginalia.nu is now open source.](https://memex.marginalia.nu/log/58-marginalia-open-source.gmi)</ins>
[Teclis](http://teclis.com/)
: A project by the creator of Kagi search. Uses its own crawler that measures content blocked by uBlock Origin, and extracts content with the open-source article scrapers Trafilatura and Readability.js. This is quite an interesting approach: tracking blocked elements discourages tracking and advertising; using Trafilatura and Readability.js encourages the use of semantic HTML and Semantic Web standards such as [microformats](https://microformats.org/), [microdata](https://html.spec.whatwg.org/multipage/microdata.html), and [RDFa](https://www.w3.org/TR/rdfa-primer/). It claims to also use some results from Marginalia. This has been down for maintenance for a while; I may end up having to move it to the Graveyard section.
: A project by the creator of Kagi search. Uses its own crawler that measures content blocked by uBlock Origin, and extracts content with the open-source article scrapers Trafilatura and Readability.js. This is quite an interesting approach: tracking blocked elements discourages tracking and advertising; using Trafilatura and Readability.js encourages the use of semantic HTML and Semantic Web standards such as [microformats](https://microformats.org/), [microdata](https://html.spec.whatwg.org/multipage/microdata.html), and [RDFa](https://www.w3.org/TR/rdfa-primer/). It claims to also use some results from Marginalia. [The Web interface has been shut down](https://kagifeedback.org/d/1838-teclis-is-broken/2), but its standalone API is still available for Kagi customers.
### Site finders
@ -240,7 +239,7 @@ These engines try to find a website, typically at the domain-name level. They do
: The best in this category. Has a small but growing index of over 8 million sites. If I want to find the website for a certain project, Kozmonavt works well (provided its index has crawled said website). It works poorly for learning things and finding general information. I cannot recommend it for anything serious since it lacks contact information, a privacy policy, or any other information about the org/people who made it. Discovered in the seirdy.one access logs.
[search.tl](http://www.search.tl/)
: Generalist search for one <abbr title="top-level domain">TLD</abbr> at a time (defaults to .com). I'm not sure why you'd want to always limit your searches to a single TLD, but now you can.[^10] There isn't any visible UI for changing the TLD for available results; you need to add/change the `tld` URL parameter. For example, to search .org sites, append `&tld=org` to the URL. It seems to be connected to [Amidalla](http://www.amidalla.de/). Amidalla allows users to manually add URLs to its index and directory; I have yet to see if doing so impacts search.tl results.
: Generalist search for one <abbr title="top-level domain">TLD</abbr> at a time (defaults to .com). I'm not sure why you'd want to always limit your searches to a single TLD, but now you can.[^11] There isn't any visible UI for changing the TLD for available results; you need to add/change the `tld` URL parameter. For example, to search .org sites, append `&tld=org` to the URL. It seems to be connected to [Amidalla](http://www.amidalla.de/). Amidalla allows users to manually add URLs to its index and directory; I have yet to see if doing so impacts search.tl results.
[Thunderstone](https://search.thunderstone.com/)
: A combined website catalog and search engine that focuses on categorization. Its [about page](https://search.thunderstone.com/texis/websearch19/about.html) claims: <q cite="https://search.thunderstone.com/texis/websearch19/about.html">We continuously survey all primary COM, NET, and ORG web-servers and distill their contents to produce this database. This is an index of _sites_ not pages. It is very good at finding companies and organizations by purpose, product, subject matter, or location. If you're trying to finding things like _'BillyBob's personal beer can page on AOL'_, try Yahoo or Dogpile.</q> This seems to be the polar opposite of the engines in the ["small or non-commercial Web" category](#small-or-non-commercial-web).
@ -358,6 +357,9 @@ Some search engines are integrated into other appliances, but don't have a web p
These engines were originally included in the article, but have since been discontinued.
[Petal Search](https://web.archive.org/web/20230118225122/https://www.petalsearch.com/)
: A search engine by Huawei that recently switched from searching for Android apps to general search in order to reduce dependence on Western search providers. Despite its surprisingly good results, I wouldn't recommend it due to privacy concerns: its privacy policy describes advanced fingerprinting metrics, and it doesn't work without JavaScript. Requires an account to submit sites. I discovered this via my access logs. Be aware that in some jurisdictions, it doesn't use its own index: in Russia and some EU regions it uses Yandex and Qwant, respectively. Inaccessible as of June 2023.
[Neeva](https://web.archive.org/web/20230528051432/https://neeva.com/blog/may-announcement)
: Formerly in [the "semi-independent" section](#semi-independent-indexes). Combined Bing results with results from its own index. Bing normally isn't okay with this, but Neeva was one of few exceptions. Results were mostly identical to Bing, but original links not found by Bing frequently popped up. Long-tail and esoteric queries were less likely to feature original results. Required signing up with an email address or OAuth to use, and offered a paid tier with additional benefits. Acquired by Snowflake and announced its shut-down in May 2023.
@ -393,7 +395,7 @@ Why bother using non-mainstream search engines?
### Conflicts of interest
Google, Microsoft (the company behind Bing), and Yandex aren't just search engine companies; they're content and ad companies as well. For example, Google hosts video content on YouTube and Microsoft hosts social media content on LinkedIn. This gives these companies a powerful incentive to prioritize their own content. They are able to do so even if they claim that they treat their own content the same as any other: since they have complete access to their search engines' inner workings, they can tailor their content pages to better fit their algorithms and tailor their algorithms to work well on their own content. They can also index their own content without limitations but throttle indexing for other crawlers.[^11]
Google, Microsoft (the company behind Bing), and Yandex aren't just search engine companies; they're content and ad companies as well. For example, Google hosts video content on YouTube and Microsoft hosts social media content on LinkedIn. This gives these companies a powerful incentive to prioritize their own content. They are able to do so even if they claim that they treat their own content the same as any other: since they have complete access to their search engines' inner workings, they can tailor their content pages to better fit their algorithms and tailor their algorithms to work well on their own content. They can also index their own content without limitations but throttle indexing for other crawlers.[^12]
One way to avoid this conflict of interest is to _use search engines that aren't linked to major content providers;_ i.e., use engines with their own independent indexes.
@ -504,8 +506,10 @@ Some of this content came from the [Search Engine Map](https://www.searchenginem
[^9]: More information can be found in [this HN subthread](https://news.ycombinator.com/item?id=27593801) and some posts on the Cliqz tech blog ([one](https://0x65.dev/blog/2019-12-06/building-a-search-engine-from-scratch.html), [two](https://0x65.dev/blog/2019-12-10/search-quality-at-cliqz.html)).
[^10]: Some search engines support the `site:` search operator to limit searches to subpages or subdomains of a single site or TLD. `site:.one`, for instance, limits searches to websites with the ".one" TLD.
[^10]: I will explain my thinking in another post later, and then edit this with a link to that post.
[^11]: Matt from Gigablast told me that indexing YouTube or LinkedIn will get you blocked if you aren't Google or Microsoft. I imagine that you could do so by getting special permission if you're a megacorporation.
[^11]: Some search engines support the `site:` search operator to limit searches to subpages or subdomains of a single site or TLD. `site:.one`, for instance, limits searches to websites with the ".one" TLD.
[^12]: Matt from Gigablast told me that indexing YouTube or LinkedIn will get you blocked if you aren't Google or Microsoft. I imagine that you could do so by getting special permission if you're a megacorporation.