Weighted Ensemble LSTM Model with Word Embedding Attention for E-Commerce Product Recommendation

Published online: Dec 12, 2023 Full Text: PDF (1.32 MiB) DOI: https://doi.org/10.24138/jcomss-2023-0126
Cite this paper
Prashant Sharma, Vijaya Ravindra Sagvekar


Nowadays, the proliferation of social media and e-commerce platforms is largely due to the development of internet technology. Additionally, consumers are used to the idea of using these platforms to share their thoughts and feelings with others through text or multimedia data. However, it is difficult to identify the best categorization methods for this type of data. Furthermore, users are seen to have difficulty understanding aspect-based feelings conveyed by other users, and the currently existing models’ accuracies are not up to par. Deep learning models used for sentiment analysis (SA) provide improved performance by finding out the actual emotions in the presented data. The aim of this research is to develop a weighted ensemble with Long Short-Term Memory (LSTM), and a specialised deep learning model using unique word embedding approaches to enhance its performance in sentiment analysis. The words with a strong connection to a particular class are given more weight by the Word Embedding Attention (WEA) technique. The weighted ensemble with LSTM yields superior outcomes because of its excellent generalization capabilities. By integrating the advantages of several models and mitigating the effects of each model’s shortcomings, ensemble voting raises the prediction accuracy. By lessening the influence of outliers or errors in individual model categorization, ensemble voting increases the robustness of categorization. This LSTM weighted ensemble achieves 99.82 % accuracy, 99.4% precision, 99.02% f-score, and 99.7% recall in sentiment analysis which is much higher when compared to the outcomes of conventional methods.


Deep learning, Long Short-Term Memory, Sentiment analysis, Weighted Ensemble, Word embedding attention
Creative Commons License 4.0
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.