Cross-Platform Generalization in E-Commerce App Sentiment Analysis: A Large-Scale Comparative Study of Classical, Recurrent, and Transformer Architectures

DOI

10.2991/978-94-6239-697-5_17How to use a DOI?

Keywords

Sentiment analysis; E-commerce reviews; Cross-platform generalization; LSTM; GRU; RoBERTa; TF-IDF; App store reviews; Domain adaptation

Abstract

Sentiment analysis of mobile app reviews presents a challenging natural language processing problem owing to the short, noisy, and linguistically diverse nature of user-generated content. In this paper, we present a large-scale comparative study on approximately 1.35 million English-language reviews scraped from the Google Play Store listings of Amazon India and Flipkart, two of the largest e-commerce platforms in India. We evaluate six classical machine learning models using TF-IDF features alongside four recurrent deep learning architectures LSTM, BiLSTM, GRU, and BiGRU, on binary sentiment classification (positive vs. negative), using Macro F1 as the primary evaluation metric. To investigate cross-platform generalization, models trained on one platform are evaluated on the other without fine-tuning, and additionally bench-marked against a fine-tuned RoBERTa transformer. Our results show that Linear SVM achieves a competitive Macro F1 of 0.9011, approaching the best deep learning model (GRU: 0.9033) at a fraction of the computational cost. Cross-platform transfer reveals a pronounced asymmetry: models trained on Flipkart’s data generalize effectively to Amazon’s data (GRU ∆ = +0.0019), whereas Amazon-trained models suffer significant degradation on Flipkart (GRU ∆ =  −0.0679). This asymmetry is consistent across all three model families including RoBERTa, demonstrating that source-domain class imbalance, rather than model capacity or pretraining strategy constitutes the primary bottleneck in e-commerce cross-platform sentiment transfer.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Yash Kumar Arora
AU  - Karan Verma
AU  - Akshay Singh
PY  - 2026
DA  - 2026/06/04
TI  - Cross-Platform Generalization in E-Commerce App Sentiment Analysis: A Large-Scale Comparative Study of Classical, Recurrent, and Transformer Architectures
BT  - Proceedings of the Conference on Bridging Engineering Disciplines with AI and Machine Learning (BEDAIML 2026)
PB  - Atlantis Press
SP  - 195
EP  - 209
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-697-5_17
DO  - 10.2991/978-94-6239-697-5_17
ID  - Arora2026
ER  -