Diabetic Food Recommendation System Using LSTM with Self-Attention and Text GAN for Enhanced Novelty and Relevance

• Satish Singh Mekale,
• Maumita Chakraborty,
• Chiradeep Mukherjee,

Vol. 19, No. 11, November 30, 2025
10.3837/tiis.2025.11.002, Download Paper (Free):

• Diabetes Mellitus
• Text GAN
• LSTM
• self-attention
• Food Recommendations

Abstract

Personalised meal recommendation systems are becoming progressively more useful for those with unique dietary needs, especially diabetics who must carefully plan their meals. This paper explores the potential of deep learning models to generate meal suggestions that are personalized to each individual and accommodate their circumstances. We experimented and compared several models, including LSTM, GRU, Transformer, LSTM with a self-attention mechanism, and TextGAN, to determine how well they realized dietary requirements and provided meal suggestions in accordance with those requirements. The suggested approach includes pre-processing data, using LSTM to encode patient and food attributes, utilising a twofold self-attention mechanism to find important nutritional limits, using TextGAN to provide a variety of recommendations, and lastly decoding to generate user-friendly meal recommendations. We employed typical natural language processing metrics to determine how well the models performed. These measures consider how well the recommendations aligned with the user's input language-wise and meaning-wise. The LSTM with a self-attention mechanism performed better than the remaining tested models, achieving perfect scores (1.0000) on all three measures. This indicates that it is more effective at grasping user preferences and dietary context. Text Generative Adversarial Networks (GAN) also performed much better with hyper parameter tuning. As an example, BLEU improved from 0.72-0.80, and ROUGE-1 and ROUGE-L also had comparable improvements. In conclusion, all these results indicate Attention-based models with generative architectures show strong potential for accurate, diverse, and personalized dietary recommendations. This effort paves the way for creating intelligent dietary aids that can assist individuals with diabetes in maintaining better health.

Statistics

Cite this article