Online Review-based Product Recommendation Using Mellivora Heron Optimization-based BiLSTM Model in E-Commerce Systems

Recently, a recommender system retrieves patterns from online reviews with the user's past interactions and provides future recommendations for users to purchase as well as consume with the upsurge development of Internet technology. By conducting sentiment analysis on a large number of user reviews on e-commerce platforms, user satisfaction can be successfully increased. However, numerous existing literature fail to extract certain implicit information and face difficulties in predicting the precise sentiment polarities of the user reviews because of the variations in sequence length, textual order, and complicated logic. To address the drawbacks of the existing techniques, the proposed research exploited the Mellivora Heron Optimization-based BiLSTM (MHO-based BiLSTM) model for product recommendation with higher quality. By incorporating the semantic ontology graph-based feature extraction involving Euclidean distance, cosine similarity, and word embeddings, along with graph embedding assists in enhancing the recommendation system. Specifically, the proposed Mellivora Heron Optimization algorithm optimally tunes the BiLSTM model to capture the context information efficiently and boosts the classifier accuracy. Moreover, a highly accurate and pertinent recommendation system for diverse scenarios is produced by combining the Spark architecture, Semantic Ontology Graph, BiLSTM classifier, and cutting-edge algorithm. With 80% of training, the MHO-based BiLSTM model for the review-based recommendation system utilizing the Amazon Product dataset obtained the accuracy, sensitivity, and specificity values of 95.29%, 95.77%, and 94.81% and outperformed other existing techniques. Similar results were found for the Amazon Video Games dataset, with values achieved during 80% of training are 96.23%, 97.20%, and 95.25%. Meanwhile, the MHO-based BiLSTM model attained the values of 96.73%, 97.71%, and 95.57%, respectively for the Amazon Video Games dataset for k-fold 10 analysis.