A Hybrid Multi-Stage Framework Integrating ARIMA-LSTM Forecasting and Transformer-Based Sentiment Analytics

Abstract

E-commerce services are characterized by dynamic prices and a large number of users' reviews - obstacles for well-informed purchase decision. In this work, we propose a new hybrid model that combines ARIMA-LSTM forecasting with Transformer-based Sentiment Analysis to provide predictive and sentiment-aware price analyses. Price lists and customers reviews are collected through ethical web scraping and pre-processed as input to analysis. The combined ARIMA-LSTM model captures both the linear and non-linear structure in data, whereas Transformer can remove context and aspect aspects from customer feedback, such as mild nuances of sarcasm and long distance dependencies. An algorithm for multi-criteria decision-making integrates forecasting and sentiment analysis to provide recommendations to platforms as to good price evolutions. Experiments on real-world data demonstrate that the proposed approach improves forecasting and sentiment analysis, and enables actionable insights from predictive recommendations. The modular architecture of hybrid model can be utilized in different retail or finance applications