User-Recipe Embedding Alignment for Dietary Preference Recommendations

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current online concierge systems face challenges in efficiently recommending recipes to users due to increased time spent browsing irrelevant recipes, which decreases user interaction and system performance.

Innovation Solution

An online concierge system uses a machine learning recommendation model with a two-tower architecture to align user and recipe embeddings in a common latent space, trained on user interactions and recipe attributes, including dietary preferences, to recommend relevant recipes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the online concierge system displays more recipes to users, then the variety of items available for selection increases, but the time users spend browsing irrelevant recipes increases

Engineering Contradiction:

Improvevariety of items availableVSAvoidbrowsing time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs preliminary actions by training a machine learning recommendation model that analyzes user interactions and recipe attributes beforehand. This model pre-computes relevance scores and embeddings for users and recipes, enabling the system to quickly retrieve and display relevant recipes without requiring users to browse through all available recipes manually.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces the mechanical browsing system with an intelligent recommendation system. Instead of relying on users to manually navigate and filter through recipes based on dietary preferences and restrictions, the system uses machine learning models to automatically match users with relevant recipes based on their profiles and interactions, substituting manual browsing with automated intelligent retrieval.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Adaptability or versatility

If the online concierge system displays more recipes to users, then the range of items for ordering increases, but system performance decreases due to increased computer processing resources and network bandwidth

Engineering Contradiction:

Improverange of itemsVSAvoidsystem performance

Core Design Contradiction:

Adaptability or versatilityVSProductivity

Solution Approach 1:

The system segments the large recipe database into manageable components by organizing recipes into sections and using a two-tower architecture that separately processes user profiles and recipe attributes. This segmentation allows the system to handle large datasets efficiently by processing and storing embeddings in distributed structures, reducing the computational burden on any single component.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary processing by pre-computing and storing embeddings for users and recipes during training phases. This preliminary action transforms the raw data into optimized representations that can be quickly retrieved and compared during runtime, significantly reducing the computational resources and network bandwidth required when users interact with the system.

Inventive Principle:

Principle #10Preliminary action

3Ease of operation

If the online concierge system displays recipes without filtering by dietary preferences, then all recipes are available for selection, but user interaction decreases due to irrelevant recommendations

Engineering Contradiction:

Improverecipe availabilityVSAvoiduser interaction

Core Design Contradiction:

Ease of operationVSReliability

Solution Approach 1:

The patent replaces manual filtering mechanisms with an automated machine learning-based recommendation system. The system automatically analyzes user profiles, dietary preferences, and recipe attributes to generate personalized recommendations, eliminating the need for users to manually filter recipes while ensuring high relevance to user needs and maintaining strong user interaction.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system implements feedback mechanisms by continuously learning from user interactions with recipes. The machine learning model updates user profiles and preference models based on observed behavior, such as recipes users select or skip, enabling the system to improve recommendation accuracy over time and maintain high user interaction levels.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20250356408A1Method, computer program product, and system for training a machine learning model to generate user embeddings and recipe embeddings in a common latent space for recommending one or more recipes to a user

Publication Date: 2025.11.20 MAPLEBEAR INC

AI summary

An online concierge system generates recipe embeddings for recipes including multiple items and user embeddings for users, with the recipe embeddings and user embeddings in a common latent space. To generate the user embeddings and the recipe embeddings, a model includes separate layers for a user model outputting user embeddings and for a recipe model outputting recipe embeddings. When training the model, a weight matrix generates a predicted dietary preference type for a user embedding and for a recipe embedding and adjusts the user model or the recipe model based on differences between the predicted dietary preference type and a dietary preference type applied to the user embedding and to the recipe embedding. Additionally cross-modal layers generate a predicted user embedding from a recipe embedding and generate a predicted recipe embedding from a user embedding that are used to further refine the user model and the recipe model.