Topic: Updated Post Suggester?

Posted under e621 Tools and Applications

Back when Material e621 still worked, it had a post suggester that used your favorites to suggest posts that were similar. Is there any tool like that now? I used it a lot to find new artists and stuff that I liked. Anything that mimics the way it works would be fine for me.

I don't think that a tool like that exists at the moment.
It does not sound too complicated to implement. I might add it to re621. We'll see.

I was about to make a post asking basically the same thing. A suggested posts feature would be really great.

Or, a way to cross reference your favorites with other users' favorites. So you could find other users that you have a lot of favorites in common with.

Anything you could do bitWolfy would be greatly appreciated! Thanks so much!

naughtynelly said: Anything you could do bitWolfy would be greatly appreciated! Thanks so much!

My problem is that I'm not satisfied with just "anything". I want to implement it well.
I've never used the material e621 site, so I have absolutely no idea how its post suggester feature worked.

Getting the favorites and tallying up the tags is easy enough: https://i.imgur.com/ixXo6d8.png
The script would check the first 10 non-generic tags - you can customize that, of course - and pressing the search button will give you a random selection of post with those.
I'm not sure if I like this approach, though.

Any insight into what exactly the post suggester feature of the material e621 did would be appreciated.

bitwolfy said:
My problem is that I'm not satisfied with just "anything". I want to implement it well.
I've never used the material e621 site, so I have absolutely no idea how its post suggester feature worked.

Getting the favorites and tallying up the tags is easy enough: https://i.imgur.com/ixXo6d8.png
The script would check the first 10 non-generic tags - you can customize that, of course - and pressing the search button will give you a random selection of post with those.
I'm not sure if I like this approach, though.

Any insight into what exactly the post suggester feature of the material e621 did would be appreciated.

As someone who used the site extensively, I can give a description of how it worked. Not in-depth code-wise, but I can describe what it was like for the end user.
You would input your name, and it retrieved your favorites, I believe it retrieved up to 1280, presumably your most recent, and then you were offered sliders for each category of tags, eg. generic tags, species tags, character tags, ect. and it would let you adjust the weight of each type of tag. eg. if you increased the Character slider, it would increase how much the character tags mattered for the suggestions. It would then open a pseudo-search with suggested posts based on your favorites and the weights of each category.

cherenthecat said:
As someone who used the site extensively, I can give a description of how it worked. Not in-depth code-wise, but I can describe what it was like for the end user.
You would input your name, and it retrieved your favorites, I believe it retrieved up to 1280, presumably your most recent, and then you were offered sliders for each category of tags, eg. generic tags, species tags, character tags, ect. and it would let you adjust the weight of each type of tag. eg. if you increased the Character slider, it would increase how much the character tags mattered for the suggestions. It would then open a pseudo-search with suggested posts based on your favorites and the weights of each category.

Favorites were fetched with the query "fav:{username}" and tags were extracted. The tags would get ordered by how often they occurred within the 4*320 most recent favorite posts and grouped by category.

As long as there weren't enough posts to fill the page, new posts (with no tags in the query, ordered by date, newest first) would be fetched and filtered (blacklist). Each post would get scored and if it reached a certain threshold, added to the page.

Tags were categorized in 5 categories (general, artist, copyright, species) and weighted by how often they appear in the user's favorites.

If a post tag was a favorite tag, the category score would increase by 0.5 + (1 * percentage of posts in the user's favorites that had the same tag). At the end, each category score was divided by the total count of tags on the post within the category. This score was multiplied by the user-supplied weight for each category.

Rating was handled separately: 4 * percentage of posts in the user's favorites that had the same rating. Those 5 category and the rating score were added and made up the post score.

Category weights were between 0 and 1 and the sum of the weights was 1.
The slider for minimum score could be adjusted between 20 and 75.

cherenthecat said:
Back when Material e621 still worked, it had a post suggester that used your favorites to suggest posts that were similar. Is there any tool like that now? I used it a lot to find new artists and stuff that I liked. Anything that mimics the way it works would be fine for me.

Someone had at least been working on a patch for Material E621 to get it working with the new API.

https://github.com/avoonix/material-e621

I'm working on building a post suggestion API, which could be integrated into a web browser as an extension to add suggested posts to e621.net.

It'll feature an option for the original post suggestion algorithm (implemented in material e621), as well as one I have come up with (involving dynamic indexes to identify trends in people that like similar content and using that for predictive suggestions).

As it's still in early development, it's not fully working yet. However, I wanted to post about it to "get the word out" so to speak, and figure out what features people wanted to see so I can plan for them (if/when possible).

If enough people are interested, I can post a Discord server link where you will be able to follow updates.

------------------------------------------ IN GERMAN/AUF DEUTSCH ------------------------------------------
Ich arbeite an der Entwicklung einer API für Beitragsvorschläge, die als Erweiterung in einen Webbrowser integriert werden kann, um vorgeschlagene Beiträge zu e621.net hinzuzufügen.

Es wird eine Option für den ursprünglichen Beitragsvorschlags-Algorithmus (implementiert im Material e621) enthalten, sowie eine, die ich mir ausgedacht habe (mit dynamischen Indizes, um Trends bei Leuten zu erkennen, die ähnliche Inhalte mögen, und diese für prädiktive Vorschläge zu nutzen).

Da es sich noch in einer frühen Entwicklungsphase befindet, ist es noch nicht voll funktionsfähig. Ich wollte jedoch darüber berichten, um es sozusagen "bekannt zu machen" und herauszufinden, welche Funktionen sich die Leute wünschen, damit ich sie einplanen kann (falls/falls möglich).

Wenn genug Leute interessiert sind, kann ich einen Discord-Server-Link posten, über den man Updates verfolgen kann.

EDIT: To avoid spamming this, I will post this in its own thread in a day or two.

Updated

  • 1