What is the science of inspiration?

Pinterest’s mission is to provide everyone with the inspiration to create a life they love, and Machine Learning sits at the core of making that mission a reality.

Machine Learning is not only powering predictive results for people across interests to help them find their next great idea, it’s identifying misinformation, surfacing relevant ads, and advancing inclusivity and body positivity, to name a few areas. Pinterest engineers are building with Machine Learning to make Pinterest representative of the people who use it every day, forging the way for a positive corner of the internet we can all be proud of, and find useful.

Watch the video or view the audio described video here.

María Luisa worked as an innovation consultant until she suffered a severe brain stroke that caused an inability to formulate language, a dysfunction known as aphasia.

After rehab, María Luisa took it upon herself to continue rehab on her own. In a moment of inspiration, she thought Pinterest might help her discover exercises to teach herself how to speak again.

Read on to see how Pinterest helped inspire the formation of María Luisa’s own project to help others experiencing the same condition.

Watch the video of María Luisa and her husband or read the transcript here.

To find inspiration for her aphasia recovery, María Luisa searches for "aphasia" on Pinterest to discover exercises.

For this initial Pinterest interaction, María Luisa begins with a query.

She starts her journey with just "aphasia" or relevant queries like "aphasia exercises" or "aphasia recovery."

We use Machine Learning to understand what the search query means and what each piece of content that populates is about—in this case, how does it relate to aphasia? The initial Pins depend on the query.

María Luisa starts with just "aphasia" and many different Pins come up since it's a simple but general term.

Results may include peripheral categories like "aphasia treatment," "aphasia quotes," "aphasia information," or "aphasia therapy."

The featured technology that allows us to present these similar results is AutoML, a ranking model framework that is applied to various personalization tasks.

Featured Tech: AutoML
AutoML is a ranking model framework that is applied to various personalization tasks (homefeed ranking, related Pins ranking, search ranking and various ads ranking). Read more

In this example, it includes expanding the query to other similar queries (i.e. aphasia -> [aphasia treatment, aphasia exercise, etc.]).

Featured Tech: Interest Taxonomy
A taxonomy-based knowledge management system that enables content understanding in a highly efficient way. Read more

María Luisa is looking for ways to enhance her therapy exercises, so she clicks on several different "aphasia therapy" and "aphasia exercise" pins.

Each user interaction (save, close up, click, etc.) with the Pin helps our Machine Learning system understand more about the Pin that is saved, and uses that to recommend additional relevant content.

As María Luisa scrolls through her home feed, she finds more and more inspiration for her situation, while also forming a more concrete idea of what she's looking for.

These related Pins inspire María Luisa to find her next idea. We help her find exactly what she's looking for—before she even knows what it is—because of the actions she takes while she browses.

Featured Tech: PinSage
A new graph convolutional neural network for web-scale recommender systems. Read more

Machine Learning refines the topic that the Pin is about and produces better quality embedding for the Pin. At the same time, it will also help refine our personalization models to help it learn that other users similar to María would like the same piece of content as well.

On the backend, we see the models shifting through 300B+ Pins to find aphasia Pins and narrowing them down to the top 10,000 "candidates" to serve to María Luisa. This is done by matching on the topic, text, or even using Machine Learning to learn the "embedding" of the query to match with the "embedding" of the candidate.

Featured Tech: PinnerSage
User representation learning based on clustering algorithms. Read more

Through Pinterest, María Luisa is able to find a lot of great ideas for aphasia exercises, flashcards, and more, but also sees the opportunity to inspire others by adding her own knowledge gained during therapy.

Now María Luisa starts creating Boards on Pinterest to educate others based on what she learned in rehab, populating them with her own flashcards from therapy to help others also seeking inspiration during aphasia recovery.

The aphasia flashcards that María Luisa creates on Pinterest become popular with others who are also searching for aphasia resources.

We see search queries of others with aphasia while María Luisa's Pins and Boards appear in the searches as well.

More and more people discover María Luisa's aphasia flashcards—and other helpful resources—because of the recommendation system and ranking.

Featured Tech: Pixie
Used in homefeed and related pins in the "candidate generation." Read more

We use Machine Learning to learn how to rank the best candidates by taking into account the users' past interactions.

Through the high engagement of her created pins, María Luisa recognized that there are a lot of people suffering from aphasia, but there is so little information about the condition.

So she decided to start her own project, called "Hola Que Tal, Afasia" ("Hi, How Are You, Aphasia"), to help educate others suffering from this disability.

María Luisa's aphasia Pinterest board can be used by speech therapists for rehab. These boards can also be helpful to relatives of people suffering from aphasia to help them practice. María Luisa also uses them in weekly workshops at her former rehab center, where she now helps others with their aphasia.

We encourage users to share their created Pins and inspire other Pinners. Users who find María's aphasia flashcards can make Story Pins of themselves trying them out, sharing their own progress, and recovery from aphasia.