Multimodal Sentiment Analysis in Quick Commerce: LSTM Networks for Text, Image, Video Feedback in FMCG Platforms

Abstract

This research explores the application of Long Short-Term Memory (LSTM) networks for performing sentiment analysis on customer reviews gathered from prominent FMCG e-commerce platforms such as Blinkit, Zepto, and JioMart. In the fiercely competitive landscape of online retail, accurately interpreting customer sentiment is essential for sustaining customer satisfaction and achieving strategic growth. These platforms accumulate massive amounts of unstructured data—ranging from feedback on product quality to delivery efficiency and overall user experience—which are challenging to process using traditional manual methods. To address this, the study leverages advanced Natural Language Processing (NLP) techniques, with a particular focus on LSTM networks due to their superior ability to model sequential dependencies and retain contextual meaning across review texts. To further enhance performance, pretrained word embeddings are used, enabling the model to understand nuanced language and improve accuracy across varying review structures. Beyond analyzing textual data, the research also integrates visual components into a multimodal sentiment classification framework, offering a holistic understanding of consumer emotions. This dual-modality approach captures subtle sentiments that may not be evident in text alone. The findings yield practical insights for enhancing customer service, optimizing product selections, and improving overall brand engagement. Ultimately, this study empowers data driven strategies that elevate user experience and market responsiveness in the dynamic FMCG e-commerce industry.