Allergy-Aware Recipe Recommender System Using NLP-Based Ingredient Matching and Allergen Filtering
Year: 2026
Type: conference-paper
Abstract: Individuals with food allergies face significant challenges when searching for safe recipes on existing platforms, which rely on keyword-based search without personalized health filtering. This paper presents an NLP-based allergy-aware recipe recommender integrating text preprocessing, TF-IDF vectorization, cosine similarity matching, and rule-based allergen filtering into a unified pipeline. The system operates over approximately 64,000 real-world recipes and accepts two user inputs - available ingredients and known allergens - to generate a ranked list of safe, relevant dishes. Spell correction and fuzzy matching align user vocabulary with dataset terms, while allergen filtering uses a manually constructed synonym dictionary covering eight common allergen categories. As a representational validity check, a Linear SVM classifier trained on TF-IDF ingredient features achieved 99.67% accuracy in safe/unsafe recipe classification, confirming that the feature space carries strong discriminative signal for allergen detection. Retrieval evaluation on 200 synthetic queries yielded a Hit@5 of 0.20 and a 0.00 allergy-violation rate across 1,000 evaluated recommendations, confirming that the allergen filter performs reliably within its defined allergen vocabulary. The system reduces the cognitive burden on users who must otherwise inspect ingredient lists manually, contributing to safer and more accessible human-computer interaction in everyday dietary management.
Institution: International Burch University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225795124