Sentiment Analysis of E-Commerce User Reviews and Prediction of Consumer Decision-Making Based on Deep Learning

Abstract

With the exponential growth of e commerce user reviews, automatically extracting sentiment information from unstructured text to predict consumer decision tendencies has become a critical yet challenging task. Existing methods either rely on shallow lexical features with limited contextual understanding or employ complex deep learning models that lack interpretability.To address these issues, this paper proposes an Emotion Driven Consumer Decision Tendency Prediction Model (ECD TTM) based on BERT and XGBoost.The model first extracts 768 dimensional sentence level semantic features from each review using the pre trained BERT model.These deep embeddings are then concatenated with a structured feature—the review helpfulness ratio—and fed into an XGBoost classifier to predict binary consumer decisions (positive vs.negative).This work offers an interpretable and effective solution for e commerce platforms to leverage review sentiment for consumer decision support.