If you use Pinterest for marketing, you know how important Pinterest search is.
The platform facilitates more than two billion searches every month, with Guided Search and Pinterests's contextual matching tools playing a big part in the visibility and reach of your Pin content - in fact, Pinterest recently reported how their upgrades to more localized search results in each nation have increased new user activity by up to 10%.
Basically, if you want to maximize Pinterest, you need to understand their search algorithm and how people are most likely to find your content. And as such, this latest update from Pinterest is important.
This week, the Pinterest engineering team has outlined how they've improved their Related Pins system to provide more relevant matches to each user.
As explained by Pinterest:
"One of the most popular ways people find ideas on Pinterest is through Related Pins, an item-to-item recommendations system that uses collaborative filtering. Previously, candidates were generated using board co-occurrence, signals from all the boards a Pin is saved to. Now, for the first time, we're applying deep learning to make Related Pins even more relevant."
And underlining the importance of on-platform search, Pinterest says that in testing, the new Related Pins system has increased engagement by 5% globally, "now accounting for nearly half of all engagement on Pinterest".
Worth paying attention to.
In an in-depth post, the Pinterest engineer Andrew Liu explains how their previous Related Pins system, using 'board co-occurrence', is often flawed because of the way people use Pinterest.
Basically, the system takes into account the types of board each image is pinned to - so if you post a picture of a dog to your 'Dogs' board, Pinterest uses that as a signal and will look to show people related pictures based on other posts in users' dogs boards.
The problem is, people use Pinterest differently.
Pinterest provides this example - in the image below, there's an image of a lion couple cuddling which has been saved to boards called 'Animals' and 'Wild animals'. As such, Pinterest uses that as context and provides related matches as shown - even though these aren't similar, in terms of image content, they are similar to the other images posted to boards alongside the first.
But that's not necessarily what people are looking for, people might actually want pictures very similar to the first, not just of other wild animals.
To solve this, Pinterest is using a new search system called Pin2Vec which uses deep learning from across the Pinterest network to not only examine related board content, but also what other users most commonly engaged with immediately after viewing the first Pin - the graphic below shows how user activity is clustered around related Pin entities.
Based on this, the results are much more aligned with wider user behavior - for example, when you search for the same picture of lions in Pin2Vec, the system provides this result.
This shows that Pinners more commonly search for exact matches for this Pin, as opposed to related images based on board matches.
In another example, Pin2Vec's Related recommendations for a Pin of a bottle of wine are actually drinks made with wine, not other wine Pins, showing that users more commonly search for related uses as opposed to the same item.
The system effectively uses audience trends and habits to highlight more contextually relevant matches, helping to guide users to more of the Pin content they'll likely be interested in.
For Pinterest marketers, this could be significant - under the previous system your products would show up based on related board matches, so people looking up your products would be shown similar product options which were not as exact. This new system could ensure they see more directly comparable products, and could help them uncover the best deal by showing which similar items other users went on to purchase.
The change could also lessen your product exposure, as improved pin matching could mean users see fewer related Pins that come from your business with each result, and are shown a wider variety from other sources.
It's difficult to know exactly how much of an impact the change will have, as it'll be a case-by-case proposition, but it's important to understand such changes and how they could impact on your platform performance.