Real-time Search, Session, and Similar Ranking

Today we’re excited to announce the release of several APIs and features that help you get the most from your Shaped ranking and recommendation models.

Real-time Search API

This release includes a new way to retrieve your ranked items or users based on their metadata attributes — all in real-time! This allows you to get more from your Shaped ranking models by giving you and your users more control over what items you want to be ranked.

The Search API can be used in many scenarios. For example, say you’ve created a Shaped ranking model for a movie recommendation system. You could use attributes like the genre, release-date or director to create category specific recommendation carousels.

Netflix homepage showing several category and time-based filtered carousels

This release gives your users more control to discover what they’re looking for. For example, imagine you’ve created a Shaped ranking model for accommodation marketplace recommendations. You could use the Search API to filter out listings based on attributes like: availability dates, guests, price, or location.

AirBnB discovery page showing explicit metadata filters that a user may want to configure (source)

How to use the Search API

We index all categorical, numerical and binary features you provide when creating your Shaped Model. This means that if you’ve already mapped your features to use our contextual ranking models — it’ll work off-the-bat.

Below are some examples showing how you can use search with our retrieval query language. Note this works for both personalized and non-personalized requests.

1. Example: Search Predicates

2. Example: Personalized Search

For more information about what is possible with Shaped’s Search API, check out our docs here.

Real-time Session Ranking

If you’ve ever used TikTok, you’ll know how good it is at recommending personalized content that reacts to every like, comment, watch or click you make on the app. If you start interacting with cat videos within a session, TikTok is definitely going to show you more cat videos within that session. We’ve added the same technology to Shaped, allowing you to get the same reactive real-time recommendations that you see in leading social products.

As well as giving this dynamic feeling to the recommendations, session based recommendations are helpful for reaching all of your user base because they address the cold-start problem. That is, the problem of not having any user data whether they’re a new user to your platform or because your traffic is anonymous. Using Shaped’s session based ranking, you just need to provide the first few interactions that someone makes to serve relevant, highly engaging recommendations.

If you want to see more cat videos on TikTok, interact with more cat videos! (source)

How to use Session Ranking

Session based ranking works with any of your Shaped models using the Rank API. Simply pass in the user’s previous interactions when calling rank and we’ll return the best rankings for that user and session journey.

For more information about using session ranking, check out our docs here.

Similar Items & Users

These two new endpoints allow you to retrieve similar items and users from a Shaped ranking model. You will often see something like a similar items recommendation on the side or bottom of a product or content detail page.

Similar items shown for each product within a food delivery marketplace (source)

1. Similar Items

To retrieve similar items, provide an item_id from your connected data store.

2. Similar Users

To retrieve similar users, provide a user_id from your connected data store.

For more information about Shaped’s similar endpoints, check out our docs here.

The Future

Fundamentally, Shaped helps businesses build better understandings of their users, content and products. To do this, we’ve done all the machine-learning and data engineering work necessary to pull your data, organize it and analyze it. This understanding - so far - has helped our customers build better, personalized recommendation and ranking experiences into their products and they’re/we’re pretty happy with how it’s all going. However, we believe there’s a lot more value we can provide by making this understanding more accessible with new endpoints. This release is the start of several new understanding endpoints we’ll be adding in the next few months!

Get up and running with one engineer in one sprint

Guaranteed lift within your first 30 days or your money back

100M+
Users and items
1000+
Queries per second
1B+
Requests

Related Posts

Jaime Ferrando Huertas
 | 
March 14, 2023

GPT-4: A New Milestone in Scaling Up Deep Learning

Jaime Ferrando Huertas
 | 
January 16, 2023

Microsoft vs Google - ChatGPT taking over search?

Javier Iranzo-Sanchez
 | 
September 1, 2022

10 ways AI will change products and experiences in the next 5 years