Did you miss a session from the Future of Work Summit? Head over to our Future of Work Summit on-demand library to stream.
Content curation app company Pinterest has worked hard in recent months to tailor its home feed — the most used part of the app — in order to boost engagement. Today, the growing company detailed the personalization systems now operating under the covers.
At one point, the company’s engineers relied on recency as a key determining factor for placing pins onto the home feed. In practice, that meant a newer pin would appear on the home feed before an older one. But a few months ago launched a new system it calls Pinnability that’s much more nuanced than that, Pinterest software engineer Yunsong Guo wrote in a company blog post today.
“It lacked the ability to effectively help Pinners discover Pins that really interest them, because a low-relevance Pin could very well appear before a high-relevance on,” Guo explained.
The Pinnability architecture is one of many improvements engineers and data scientists at Pinterest have made to improve the service for millions of users as the company expands internationally.
According to today’s post from Guo, in the course of improving its personalization systems, Pinterest explored an increasingly trendy kind of artificial intelligence called deep learning. The technique involves training artificial neural networks on a whole lot of data — like pictures, text, or speech — and then sending them new pieces of data and receiving inferences in response.
But it sounds like the implementation of deep learning — convolutional neural networks (CNNs) in particular — wasn’t as effective as some other common machine-learning methods.
“We experimented with multiple machine-learning models, including LR [logistical regression], GBDT [gradient-boosted decision trees], SVM [support vector machines], and CNN, and we use AUC [area under the curve] score in 10-fold cross-validation and 90/10 train-test split settings with proper model parameters for evaluation,” Guo wrote. “We observed that given a fixed feature set, the winning model always tends to be either LR or GBDT for Pinnability.”
The latest of these systems could well lead to a more personalized experience beyond just the home feed on Pinterest.
“We’ve also begun to expand the use of our Pinnability models to help improve our other products outside home feed,” Guo wrote.
Read the full blog post to learn more about how Pinterest has made inroads in home feed personalization.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn More