Food Recommendation as Language Processing (F-RLP): A Personalized and Contextual Paradigm

TL;DR

The paper addresses the challenge of scalable, personalized food recommendations amid an almost infinite class space and unbalanced data. It proposes Food Recommendation as Language Processing (F-RLP), a food-centric extension of Recommendation as Language Processing that combines a Context Generation stage, Counterfactual Generation (CFG) retraining, and a Query stage using a transformer-based recommender to produce tailored suggestions. A CFG-driven retraining regimen uses expert nutrition and user preferences to sharpen recommendations, while a Context Recognition Engine assembles contextually relevant option lists that are injected into the LLM workflow for improved relevance. Experimental framing shows F-RLP outperforms a KNN baseline and demonstrates contextual relevance and personalization, highlighting practical potential for health-aware, culturally informed dietary guidance using LLMs.

Abstract

State-of-the-art rule-based and classification-based food recommendation systems face significant challenges in becoming practical and useful. This difficulty arises primarily because most machine learning models struggle with problems characterized by an almost infinite number of classes and a limited number of samples within an unbalanced dataset. Conversely, the emergence of Large Language Models (LLMs) as recommendation engines offers a promising avenue. However, a general-purpose Recommendation as Language Processing (RLP) approach lacks the critical components necessary for effective food recommendations. To address this gap, we introduce Food Recommendation as Language Processing (F-RLP), a novel framework that offers a food-specific, tailored infrastructure. F-RLP leverages the capabilities of LLMs to maximize their potential, thereby paving the way for more accurate, personalized food recommendations.

Figures (3)

• Figure 1: RLP vs F-RLP
• Figure 2: For each category, we utilize comparison metrics to assess the degree of improvement achieved by different setting classes relative to the baseline model without Counterfactual Generation (No CFG). The results indicate a positive enhancement across all categories. The horizontal axis of the charts represents the specific item selected to gauge sensitivity levels, while the vertical axis measures the extent of the sensitivity level itself.
• Figure 3: Showcasing F-RLP's meaningful and precise answer comparison with GPT 3.5 on our text food query.