Toward Improving the Prediction Accuracy of a Product Recommendation System Based on Word Sequential Using LSTM Embedded

Abstract

The ability to predict purchases is crucial for e-commerce decision makers when making offers and suggestions to customers. In the development of recommendation models, two common problems often encountered are a lack of personalization and irrelevant recommendations. To address these issues, it is crucial to consider user history data, such as the user's interactions with previous products. This allows the model to learn user preferences from the past and generate more personalized and relevant recommendations. In this study, word2vec is used to provide rating predictions, while document context is enhanced using LSTM capture contextual understanding of product reviews. This study makes use of an actual dataset to test our model using an Amazon Review Dress. The results of our investigation demonstrate that, as 88% of the recommendations are made in accordance with the recommendation system's criteria, it can be considered that it offers reasonably accurate and dependable recommendations with an accuracy of 0.8752