Logistic Regression Model Performance.

Traditional survey-based feedback has given way to real-time online reviews, yet transforming this unstructured text into actionable knowledge remains difficult. Focusing on the highly competitive smartphone market, where customer sentiment shapes brand perception and product strategy, this study proposes an end-to-end analytics pipeline that combines Machine Learning (ML), Deep Learning (DL), topic modelling, and interactive visualisation. Reviews for ten smartphone brands from Amazon (n ≈ 68 k) were pre-processed and class-imbalanced data were mitigated through class weighting. Sentiment classification was performed with ML models (Decision Trees, Logistic Regression, SVM, Naive Bayes) and DL models (Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM)). CNN achieved the highest accuracy 85.07% and balanced performance across positive and negative classes, although all models struggled with neutral reviews. Underlying themes were extracted with Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF); quantitative evaluation using the Coherence Score (CS) showed that NMF produced more interpretable topics (CS = 0.54) than LDA (CS = 0.41). Topic-level sentiment was assessed with the Valence Aware Dictionary and Sentiment Reasoner (VADER), linking features such as battery life and camera quality to positive or negative customer attitudes. Results are delivered through an interactive Tableau dashboard that allows practitioners to track sentiment trends, drill into coherent topics, and compare brand performance. The study also discusses ethical considerations, such as potential bias from imbalanced or culturally nuanced language, and outlines future work on cross-domain generalisation and fairness auditing. Overall, the integrated pipeline demonstrates that coupling CNN-based sentiment analysis with high-coherence NMF topics provides richer, business-ready insights than sentiment analysis alone.