Catalog-Aware Language Model Recipe Suggestions for Item Finding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users face difficulties in navigating online systems to efficiently gather all items needed for multi-item content, such as recipes, especially for new users, and existing systems struggle to provide accurate and detailed multi-item content at scale.
Innovation Solution
An online concierge system utilizes a large language model (LLM) to generate personalized recipe suggestions based on user preferences and catalog data, including price changes and promotions, and presents them through a user interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a traditional user interface is used to display all items for recipes, then users can access comprehensive item information, but users experience difficulty navigating and finding all desired items efficiently
Solution Approach 1:
The system pre-processes recipe information and item relationships before user interaction. The LLM analyzes recipe requirements and pre-organizes relevant items in the catalog database, so when a user views a recipe, all necessary items are already structured and highlighted in the user interface, eliminating the need for users to search through extensive catalogs manually.
Solution Approach 2:
The LLM acts as an intermediary between the user's recipe selection and the catalog database. It translates high-level recipe concepts into specific item queries, filters, and organized presentations. The LLM processes the relationship between recipes and items, generating customized UI representations that bridge the gap between comprehensive data and user-friendly navigation.
2Measurement precision
If the system provides detailed multi-item content for each user, then content accuracy and relevance improve, but the complexity of generating and managing this content increases
Solution Approach 1:
The LLM performs self-service by automatically analyzing recipe requirements, querying the catalog database, and generating personalized item recommendations without requiring manual curation or complex programming. The system feeds recipe data and catalog information to the LLM, which autonomously processes relationships and generates appropriate content, reducing the need for manual system configuration and content management.
Solution Approach 2:
The system dynamically adjusts content parameters based on user-specific factors. The LLM modifies recipe presentations, item selections, and detailed information levels according to individual user preferences, shopping history, and context. This parameter adaptation allows high accuracy for each user without requiring separate detailed content databases for every user, as the same base data is dynamically reconfigured.
3Productivity
If the system automatically generates personalized content at large scale, then content availability and speed improve, but the quality and customization of content may deteriorate
Solution Approach 1:
The system replaces traditional mechanical content generation methods (manual curation, rule-based filtering, database queries) with an LLM-based approach. The LLM processes complex relationships between recipes and items, understands contextual nuances, and generates personalized content at scale. This substitution maintains high quality through the LLM's contextual understanding while achieving fast automated generation.
Data Source
AI summary
A language model is used to suggest content based on preferences of a user of an online system and data queried from a catalog database of the online system. The online system gathers input data including a set of recipes and user data and generates a prompt for input into the language model that includes the input data. The online system requests the language model to generate, based on the prompt input into the language model, the list of recipes for the user, wherein each recipe includes a list of ingredients. The online system selects one or more recipes from the list of recipes for presentation to the user. The online system causes a device associated with the user to display a user interface with a suggestion for the user to include, in a cart, a set of ingredients of the selected one or more recipes.