Three-Vector Embeddings for Promoted-Item CTR Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional collaborative filtering systems for online concierge systems rely on binary matrices to predict user interests, using single signals like ad clicks, which are inadequate for accurately selecting promoted items based on user interactions.
Innovation Solution
An online concierge system uses a machine learning model that generates user and item embeddings, incorporating multiple interaction signals such as purchases, cart additions, and clicks, to rank promoted items alongside organic search results, leveraging a click through rate model for selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional collaborative filtering systems use binary matrices with single signals (e.g., ad clicks) to predict user interests, then the system complexity remains low, but the accuracy of predicting user interests and selecting promoted items deteriorates
Solution Approach 1:
The patent transitions from binary matrices (0/1 values) to continuous vector embeddings (dense real-valued vectors). This dimensional transformation allows the system to represent user interests, item features, and their interactions with nuanced continuous values rather than discrete binary states, thereby improving prediction accuracy while maintaining manageable system complexity through efficient vector operations.
Solution Approach 2:
The patent combines multiple interaction signals (ad clicks, item views, cart additions, purchases) into a unified embedding representation. Instead of relying on a single binary signal, the system integrates multiple behavioral signals into composite user and item embeddings, creating a richer multi-dimensional representation that improves prediction accuracy without proportionally increasing system complexity.
2Measurement precision
If the system uses multiple interaction signals (purchases, cart additions, clicks) to generate embeddings, then the accuracy of item ranking improves, but the computational complexity and data processing requirements increase
Solution Approach 1:
The patent pre-computes and stores user embeddings, item embeddings, and their interactions in advance before the actual ranking query. By performing embedding generation and storage beforehand, the system reduces the computational burden during real-time ranking operations, allowing it to leverage multiple interaction signals for accurate ranking without excessive computational complexity during execution.
Solution Approach 2:
The patent replaces traditional mechanical collaborative filtering operations (matrix factorization, iterative optimization) with neural network-based embedding models. This substitution enables the system to process multiple interaction signals more efficiently through differentiable operations and gradient-based optimization, improving ranking accuracy while managing computational complexity through modern ML infrastructure.
3Reliability
If conventional systems rely on single signals like ad clicks, then the data processing requirements remain low, but the relevance of promoted items displayed to users deteriorates
Solution Approach 1:
The patent creates a universal embedding framework that handles multiple types of user-item interactions (clicks, views, cart additions, purchases) through a unified model architecture. This multi-functional approach allows the system to process diverse interaction signals consistently, improving the reliability of promoted item relevance while efficiently managing data processing requirements through a standardized processing pipeline.
Data Source
AI summary
An online concierge system uses a machine learning click through rate model to select promoted items based on user embeddings, item embeddings, and search query embeddings. Embeddings obtained by an embedding model may be used as inputs to the click through rate model. The embedding model may be trained using different actions to score the strength of a customer interaction with an item. For example, a customer purchasing an item may be a stronger signal than a customer placing an item in a shopping cart, which in turn may be a stronger signal than a customer clicking on an item. The online concierge system generates a ranking of candidate promoted items based on the search query and using the click through rate model. Based on the ranking, the online concierge system displays promoted items along with the organic search results to the customer.