Posts Tagged ‘feature graph’

What Is Feature Graph?

Wednesday, July 23rd, 2014 by

Feature Graph is a system that we are using to categorize every asset on TurboSquid. This new system not only provides a better search result for assets, but Feature Graph creates better listings and product placement. No longer are you subject to a small set of categories that may or may not accurately describe your model — Feature Graph has over 13,000 categories.

Improved Publishing & Search Results

Let’s start with the asset “apple.” As a homonym, this can be taken in multiple ways: is it an Apple iPhone or an apple that is a fruit?  The old ways of searching for “apple” would have resulted in products of both types – iPhones and fruits. This distinction can now be made thanks to Feature Graph.  If you want to search for apple, simply select the fruit from the drop-down menu, based on your intent. Once you have made your selection, you will only see apples of the fruit variety in the the search results– nothing else.

Another example of the benefit of Feature Graph can be demonstrated by a search for “hot dog.” This instance proves that the keywords used to describe some products can actually harm the visibility of an asset. When an artist uses “hot” and “dog” as a keyword that describes an asset, the old system would display the keyword search for “dog.”  This is clearly the wrong result, as the asset is not an animal. Feature Graph does not  rely on keywords to determine what an asset is, so it properly identifies a “hot dog” as food and places it into the correct category. The outcome is a search results page that no longer includes hot dogs in the mix when a customer or artist searches for “dog.”

There is also a hierarchy tree set up for all assets. Let’s use “clownfish” as an example; if a customer were to type in animal, sea creature, fish, coral fish, anemone fish, fishes,  tropical fish, arcanthurus, amphiprioninae, clown fish, or clownfish, they would see this asset in those lists:

This helps your product’s visibility and reduces the amount of spam and incorrectly-keyworded assets from showing up in those categories.

Product Pricing

As many of our artists have experienced, CheckMate enforces a strict pricing policy, put into place partially  to prevent undercutting. Feature Graph gives the CheckMate Inspectors a way to more accurately determine the average selling price of all assets of the same type of model. This information will provide the information needed to help TurboSquid to fend off undercutting, especially as we continue to view each model submitted against comparable assets.

Help Us Improve Feature Graph

While the number of categories continues to expand, there will likely come a time when a specific category does not have representation with an asset. This is bound to happen as we strive to categorize all of the objects in the world. We are currently developing a tool for artists to add brands to Feature Graph that aren’t yet listed  This is where you, our artists, come in: help us to better categorize your assets. Simply email us at  breadcrumbs@turbosquid.com with the product ID or URL, as well as suggestions for the specific category or brand name that you feel best represents the model. With your help, along with our own research, we can categorize every asset on TurboSquid, making it a place that allows your assets easy visibility and ultimate profitability.

Corey Cambre is a TurboSquid CheckMate Inspector.  In addition to this post, he has also detailed the process behind getting our artists’ models through the CheckMate Pro v2 Certification.

No More Overstuffed Spam Tags: The Future of Keywords in Feature Graph

Friday, October 4th, 2013 by

As artists, you’ve been given a sneak peek into our evolved categorization system, otherwise known as Feature Graph. Now that you’ve had a chance to test drive it, we’re letting you in on another secret.  The current content publishing process has allowed artists the ability to suggest keyword tags that you feel best describe your 3D creations. While this function isn’t going anywhere, we are refining how tags are processed and translated into Feature Graph assignment.

In order to fully understand Feature Graph’s keyword safety net, here are the three types of bad keywords that we’re hoping to avoid:

Lazy Tagging

lazyKWs

 

Lazy keywords happen when we forget to add the correct tags to an asset.  For instance, a car could be tagged Nissan and Skyline,  but the tag sedan is forgotten, making it difficult to assign this asset to the correct category.  Adding keywords up the category chain also helps improve assignments, so adding vehicle and car to that asset would help it get assigned correctly down the line.

 

Fluff Tagging

fluffKWs

Fluffing keywords gets in the way of good assignments.  For instance, a Dodge Charger could be tagged with coupe, vehicle, and car, but it was also tagged with filler keywords like drift, drag, rally, and racing.  A real car could be tangentially involved with drag racing, but for filing purposes, a car is just a car.

We’ve seen a lot of “free association” keywords, as well: scooters tagged with road, highway, or urban; fruit tagged with tree, seeds, plant; and food tagged with restaurant or kitchen.  All of these items and their descriptors are only tangentially related, but when you’re tagging an asset, consider what your customer is realistically going to look for.  From what we’ve observed, customer searches are quite specific.  When looking for a scooter, the customer is going to type the word scooter, or perhaps a brand name.  When they’re looking for a cake, it is unlikely that they will search for kitchen.

These filler keywords can file an asset into a category where it does not belong, causing overloaded searches (which is frustrating for customers).  Always keyword for what your model actually is, not for what it could be.

 

Spam Tagging

spamKWs

Spam keywords are similar to fluffed keywords, but in this case, the keywords are in direct competition with each other: for instance, tagging a phone with both iPhone and Blackberry.

Spam tagging is unfair to your fellow artists in the search results, and it’s bad for customers, as it causes messy search results that make it very difficult for them to find exactly what they need.

A few other tips:

Your tags should be almost always be nouns (or proper nouns).  When looking for a scooter, a customer will not type in adjectives like speedy or shiny.  

Always spell check your keywords!  Misspelling scooter as skooter will make sure that your asset does not show up in the appropriate search.  And, of course, make sure you are tagging in English.

So, what does Feature Graph’s tagging capabilities mean for you? You will still have the ability to suggest tags, however, please be aware that adding non-relevant or competing keywords may return strange or incorrect results within the Feature Graph assignment process.  Using the example above, tagging a cellphone with the terms iPhone AND Blackberry could slow the process of your item being properly categorized, and therefore would not be immediately available on the site.

A well organized site and a fast, clean search makes sure that your assets will be seen where they’re meant to be seen.  With the safety net built into Feature Graph’s keyword system, our customers will be able find exactly what they want, when they want it… and, of course, having a properly tagged catalog can eventually help us find the gaps in products, so our artists can know what to create next.

Terms of Use Privacy Policy Site Map © 2013 TurboSquid