MycGNN: enhancing recommendation diversity in e-commerce through mycelium-inspired graph neural network

This study introduces MycGNN, a novel hybrid graph neural network (GNN) inspired by the development and exploration strategies of mycelium, the vegetative structure of fungi. By dynamically reweighing edges in the recommendation stage to respond to user interactions and item features, the proposed approach tackles recommender system accuracy and diversity issues. MycGNN uses a probabilistic exploration–exploitation method to balance off exploration and exploitation, offering more personalized and diverse suggestions than traditional GNN. The proposed model is tested using Retail Rocket and MovieLens1M datasets. Its effectiveness is demonstrated in terms of both accuracy and intra-list diversity. Our results show that although state-of-the-art models may exceed MycGNN in accuracy, it outperforms them in diversity while maintaining high overall accuracy.