Hybrid self-attention BiLSTM and incentive learning-based collaborative filtering for e-commerce recommendation systems

Abstract

Sentiment prediction means analyzing the emotional tone or opinion contained in textual data such as reviews or posts on social media. Recommendation systems use this sentiment analysis to recommend appropriate products or content to the users. The limitations of the existing model are related to data quality and quantity issues and dealing with different scenes and languages. Hence, to overcome all these challenges, the Hybrid Self-Attention Layer Optimized Incentive Learning–based Collaborative Filtering-BiLSTM (Hybrid AT-IN based CF-BiLSTM) model was developed for sentiment prediction and recommendation based on e-commerce platforms. The usage of CF, BiLSTM networks, and a hybrid self-attention mechanism ensure the model's unrivaled precision and performance in the domain of sentiment analysis and consumer preferences. Relying on the CF, the model accumulates valuable data about user-item interactions, and BiLSTM networks process the text, employing information from the surrounding context. The model utilizes a hybrid self-attention mechanism that automatically assigns weights based on the importance of words in user reviews; this allows it to focus on the main features and improve understanding of sentiments. Moreover, applying incentive learning lets the model adapt and optimize recommendations based on changing user behaviors, leading to greater user satisfaction and engagement. In particular, the CCO-TLI model showcases significantly superior values with 98.03% accuracy, the lowest mean square error value of 1.42, 98.68% precision, 97.04% recall, and the smallest root mean squared error of 1.19 compared to existing models.