Hybrid Deep Learning Recommendation System for Accurate Movie and Product Review Predictions
Main Article Content
Abstract
This paper investigates the efficacy of deep learning models for sentiment analysis using two publicly available datasets: IMDb’s movie review dataset and Amazon’s product review dataset. The main objective was to evaluate the performance of various model architectures, particularly Long Short-Term Memory (LSTM) networks with dropout techniques, in emotion categorization under different settings. Key performance metrics, including accuracy, precision, recall, and F1 score, were used to train and validate several deep learning models: LSTM Spatial Dropout 1D, Bidirectional GRU-LSTM, Hybrid LSTM+GRU, and Bidirectional LSTM. The LSTM Spatial Dropout 1D model achieved remarkable results, with 93.00% accuracy and F1 score on the Amazon dataset, and an impressive 96.20% accuracy and 97.78% F1 score on the IMDb dataset. The Bidirectional GRU-LSTM model also performed exceptionally, achieving 98.69% accuracy, 96.16% precision, 94.62% recall, and 93.49% F1 score, outperforming many existing hybrid models in recommendation systems. By integrating forward and backward context, the Bidirectional GRU-LSTM model effectively captures complex temporal relationships, offering more accurate recommendations than traditional models that analyze data separately. This study underscores the robustness of LSTM-based architectures in sentiment analysis and highlights the potential of combining sentiment analysis with collaborative filtering to enhance precision and specificity in e-commerce recommendation systems.
Article Details

This work is licensed under a Creative Commons Attribution 4.0 International License.