An Emotion-Aware Recipe Generation Framework Using Distilbert and Large Language Models

Traditional food recommendation systems often provide static meal suggestions based on previous user behavior, without the ability to develop tailored recipes based on current emotional states. This study presents an emotion-aware recipe generating framework that dynamically generates personalized culinary recommendations based on user-provided emotional inputs. The system uses a pre-trained DistilBERT model for sentiment classification and keyword-based food concept mapping to match emotions to relevant food categories. These inputs are subsequently processed by Falcon-7B-Instruct, a big language model, which uses rapid engineering to generate entire, structured recipes in plain language that include both ingredients and preparation instructions. Unlike existing methods that rely on curated food databases, our approach makes use of the generative capabilities of large language models (LLMs) to provide context-sensitive, emotionally meaningful recipe recommendations. The overarching goal is to improve user engagement and fulfilment by tailoring nutritional recommendations to individual emotional states, resulting in a novel junction of affective computing and customized nutrition.