Classification and analysis with Deep Learning forPersonalized Product Recommendation in ECommerce

Abstract

Environmental e-commerce" refers to an online business model that prioritises environmental responsibility.
Environmental e-commerce could benefit from a fresh strategy that takes into account a large number of social interactions in order
to address the problems with data sparsity and variety that plague conventional e-commerce recommendation algorithms: combining
filters. This study introduces a recommendation framework based on Bidirectional Encoder Representations from Transformers
(BERT), designed to capture contextual relationships in customer reviews for accurate product suggestions. The Amazon product
review dataset was utilized, and preprocessing steps included handling missing values, tokenization, stemming, stop-word removal,
and TF-IDF. The proposed BERT model was fine-tuned on the preprocessed dataset and compared against traditional models such
as CNN, GRU and VADER. A higher accuracy rate of 89% was attained by the BERT-based method, which substantially
outperformed the comparison models, according to the experimental evaluation. These results demonstrate the effectiveness of
transformer-based architectures in understanding semantic meaning and user intent, thereby enabling more reliable and scalable
product recommendation systems in e-commerce platforms.