Here's a proposed solution for the search engine algorithm drama

Continuing the discussion from The new search engine algorithm:

This idea comes from my experience as an AJ seller. Imagine a scenario where you put the value less on search term and more on overall relevance/value of the track/file. Perhaps this is a bit easier than you’d think.

Two ways, summarized, then explained:

- Change the layout of the search results page to tempt/empower the buyer with different options and mindsets,
- Train every human approval person to give a multi-value grade, thus bettering search results.

The first part goes like this. Currently the search return page brings back a single column of results,

I see a lot of unused space to the left and right on my desktop computer (most likely where the majority of buyers go to buy/download, versus mobile browsing). Now imagine if the search results, instead of returning a single column of tracks, which currently seem loosely organized by number of sales, returned two or more ‘buckets’ or columns containing relevant content, BUT- offered in different categories different mindsets of buyers. This bucket approach ditches the pulldown sorting in favor of visual regions (thus perhaps equalizing the chances of exposure a bit more because people aren’t having to pull down to get to trending, etc).

  • One column / region is called TOP ALL TIME SELLERS. This appeals to the folks out there who want the most commonly appreciated files (at the risk of downloading relatively well-used material). These people want safe, successful work. Every seller in here has sold more than X,XXX copies over the entire life of the database.

  • Another column / region is called TRENDING SELLERS. This region of the page appeals to people looking to take a risk on something perhaps slightly more edgy but still popular. Everyone in here has sold more than a certain number of copies (in relation to the song’s go live date) over the last 60 days, period. Right now it seems that ‘trending’ often contains songs which haven’t sold any copies, which is weird. I could be mistaken though, but am pretty sure I’ve seen this.

  • Another column / section is, called DISCOVER, contains everything else. It should have a call to action that promotes the adventure basically screaming ‘find new music, be your own person, don’t follow the herd,’ that sort of general presence to it. This section appeals to the trend-setters, tastemakers, etc. And here’s where everyone who isn’t a top seller/trender gets fairly presented. This also takes us to the second main proposal of this post.

Now my idea is that every song on AJ, perhaps every Envato file, will need to be privately graded, perhaps on a maybe two or so quick criteria (musical quality, production quality) by the original approving staff member. Each criteria gets a number from 1-100 and NOONE knows what it is except the AJ staff (not even the artist). So a given song with say, 40/70/96 (for music it would mean it is an OK song with solid production quality and a very spot on relevance to its overall suggested theme and title. for other things, like visual media, those values still work), is approved. Now it has a really organic, human-friendly value assigned to it with very little extra work on behalf of the Envato staff.
So over time those values starts to create a nice range of value in the database. And, while staff members will inevitably vary in their grading, it should mostly come out in the wash. Don’t we all pretty much know when a song or image is great?

The next thing is simple: when people search “uplifting” or “ham noises” or whatever, they get a selection of music that shows up based on its keyword relevance AND its general human value behind the scenes, thus allowing actual good music to show up as it should.

One challenge is, of course, all existing Envato media. One bit of good news for existing files is that the current best sellers and so on would show up in the top selling regions on the results page. As for the rest, well, perhaps all existing files (even best sellers) gets a 75/75/75 or something, given that the best sellers will continue to show up in their section, etc. Not my forte’ at this time of night, someone needs to figure that part out fairly and correctly, or ask me on a different day. Regardless, a ‘thumbrprint rating’ could potentially be calculated and assigned for all existing titles (even top sellers) based on each item’s current trend data pretty easily. Then you just roll them into a lump and let it come out in the wash based on what people are searching. Note that this still empowers sellers to leverage keywords, but it’s in their interest to be very accurate AND make great material (as humans will not only be approving, but now also grading in an instant).

This is admittedly a late night ramble but it feels pretty good at the moment. I’ll get back to my Earl Grey now, thanks for reading and discussing.

Nice ideas! The 3-column results page is a cool concept.

Not sure about hidden grading giving certain items a better chance over others. I agree there is still a difference between a ‘good’ track and a ‘great’ track, but I trust that anything that goes past the review team is fit for selling - and thus should be in a fair playing field. I expect the quality threshold is gradually increasing anyway (which again is a good thing).

As you said, there’s a LOT of older media and going through giving each and every item a secret ‘quality rating’ could be difficult. Though I definitely like your 3-column idea - that would be a fantastic implementation.

Anyway, I’ve noticed some small tweaks here and there with the search engine (that seem to be an improvement), so fingers crossed we’ll be seeing more improvements to fix the current issues! :slight_smile:

1 Like