Consumer Product Recommendation System using Adapted PSO with Federated Learning Method

In this paper, we proposed adapted particle swarm optimization integrated federated learning-based sentiment analysis integrated deep learning (aPSO-FLSADL) model for personalized recommendations of consumer electronics that leverage SentiWordNet and BERT for word embedding, CNN-BiLSTM based Federated learning model to train a global sentiment analysis model and mutation operator based modified particle swarm optimization for learning parameter optimization in federated learning environment. SentiWordNet is a sentiment lexicon that provides sentiment scores for words, while BERT is a powerful pre-trained deep learning model for natural language processing. Our approach involves pre-processing the text data, calculating sentiment scores using SentiWordNet, converting text data into word embedding using BERT, and assigning weights to words based on a defined weighting scheme. We evaluate the performance of our approach on a separate evaluation dataset including Amazon review dataset and CNET dataset. Based on the various evaluation metrics including accuracy, loss, hit ratio, we demonstrated the effectiveness of proposed aPSO-FLSADL in generating accurate and personalized recommendations. The depicted result shows that proposed aPSO-FLSADL achieved highest training and testing accuracy for both datasets and outperform over baseline models with maximum hit ratio for consumer electronics product recommendation.