LSTM-based deep learning framework for sentiment classification of Flipkart product Reviews

The primary objective of this study is to develop a deep learning-based model for multi-class sentiment analysis of Flipkart product reviews, categorizing them into positive, negative, and neutral classes. The dataset comprises 21,904 product reviews, and data preprocessing was performed using Pandas and NumPy, including handling missing values and duplicate removal. Text preprocessing techniques such as tokenization, stop word removal, and stemming were applied using NLTK. Exploratory Data Analysis (EDA) was conducted using Matplotlib and Seaborn, while Word Cloud visualization was used to identify frequently occurring terms. The textual data was transformed into numerical sequences using the TensorFlow Keras Tokenizer and standardized through sequence padding. A deep learning model based on Long Short-Term Memory (LSTM) was implemented using TensorFlow/Keras, incorporating an Embedding layer, LSTM layer, Dropout layer, and Dense output layer with Softmax activation. The model was trained using the Adam optimizer with Early Stopping for optimization. The proposed model achieved a high-test accuracy of 96.94%, and evaluation using Scikit-learn metrics demonstrated strong performance with high precision, recall, and F1-score across all sentiment classes. The confusion matrix further confirms the model’s effectiveness in accurately classifying positive, negative, and neutral reviews. The study demonstrates that LSTM-based deep learning models, when combined with effective preprocessing and feature engineering techniques, can achieve highly accurate multi-class sentiment classification, proving robust and reliable for analyzing real-world product review data.