hybrid deep learning and evolutionary feature selection for real-time product recommendations

The swift expansion of e-commerce has driven the creation of recommendation systems (rs) that help users navigate vast catalogues and make informed purchase decisions. this work presents a novel recommendation system framework integrating adaptive techniques for enhanced accuracy and efficiency. the system utilizes adaptive evolutionary feature selection (aefs), a novel feature selection algorithm combining genetic algorithms and reinforcement learning to select the most relevant features from user interaction data, product details, and contextual data. the pre-processing stage comprises text tokenization, normalization, and stop-word removal, followed by feature extraction using term frequency-inverse document frequency (tf-idf) and latent factor modelling. user profiling is performed using graph-based profiling and behavioural profiling, allowing for a holistic view of user inclinations and preferences. the bidirectional encoder representations from transformers for recommendations (bert4rec) model, which uses transformer-based architectures, is used for generating recommendations by capturing complex sequential relationships in user behaviour. this hybrid approach combines collaborative filtering (cf) and content-based filtering (cbf) to deliver accurate and personalized recommendations. real-time recommendations are provided using a distilled model, ensuring scalability and efficiency for large-scale e-commerce platforms. the system continuously adapts through a feedback loop based on user interactions, using reinforcement learning to improve performance. with an accuracy of 98%, bert4rec achieves improvements of up to 18.45% across key metrics. the proposed framework enhances recommendation accuracy, achieves a feature reduction rate of 70%, and ensures a robust user experience in modern e-commerce environments.