Recommendation Engine Technology

Machine Learning

Fully proprietary and inspired by collaborative filtering concepts, Crossing Minds trains deep learning models using a data set of users’ ratings.

Deep Collaborative Filtering

Our Deep Collaborative Filtering approach scales to settings surpassing traditional boundaries with hidden layers and non-linear activation functions. By adding hidden layers and training with in-house proprietary sparse gradient descent, our AI can identify and learn more complicated non-linear patterns between users and items.

Semantic Graph Embedding

Semantic Graphing makes the goal of finding correlations between seemingly unrelated data simpler by using variables such as metadata, labels, tags, genre, actors and more. By doing this we can make sense of semantic data in the same way a human mind would. In a nutshell, if we know a user likes a certain item and that item shares similarities with another, we know the second item will make a good recommendation, too.

Deep Content Extraction

Deep Content Extraction allows Crossing Minds models to recommend items no one has interacted with yet. The algorithms can understand the content’s genre by automatically extracting information from items such as cover art, a synopsis, or reviews. Deep Content Extraction allows Crossing Minds models to generate accurate recommendations for new items as soon as they are available in the data set.

The Crossing Minds API

Fully proprietary and inspired by collaborative filtering concepts, Crossing Minds trains deep learning models using a data set of users’ ratings.

Truly Individualized Item Recommendations

Compared to most recommendation engines deployed today, Crossing Minds models don’t cluster individuals in buckets of hundreds of thousands of users. Instead, Crossing Minds’ recommendations are truly tailored to each profile.

Man liking an item via web, saving another item for later, and
eventually making a purchase thanks to behavior-based
recommendations

Time Decay & Taste Evolution

Compared to most recommendation engines deployed today. Instead, Crossing Minds’ recommendations are truly tailored to each profile, factoring in nuances like the relationship between time and preference.

Behavior-Based Recommendations & Database Transfer Learning

The vast majority of online browsing occurs without any long-term user identifier, hence being able to deploy behavior-based recommendations in seconds - not hours - is critical to remain relevant. Additionally, our platform is capable of on-the-fly transfer learning and predictions.

Runtime Efficiency in Production

Our API leverages a blazingly fast proprietary nearest neighbor tree index and database. Most recommendation APIs lack this combined expertise and often rely on off-the-shelf database technologies instead, limiting what they can deploy to overly simplistic models.

Query Filters

Crossing Minds’ API does not limit the available filters to only a few selected ones for a particular vertical. Instead, customers can create their own properties and attributes and modify them at any time, allowing for custom business rules to control the recommendations.

Platform Integrations

The Crossing Minds solution integrates with key CDPs and PaaS, including but not limited to Segment, mParticle, and BigQuery. With just a few clicks, customers can feed their users’ events directly into the Crossing Minds API without writing any code.

Realtime Training

Crossing Minds' API automatically re-trains the recommendation following event-based triggers. This means that each and every time your database changes significantly, the entire model is re-trained from scratch.

Recommendation Engine Technology Powering Crossing Minds