Decoding brand sentiments: Leveraging customer reviews for insightful brand perception analysis using natural...

Area of Science:

• Data Science
• Artificial Intelligence
• Natural Language Processing

Background:

• Traditional feedback methods are insufficient for real-time online reviews.
• Analyzing customer sentiment is crucial for brand perception and product strategy in competitive markets like smartphones.

Purpose of the Study:

• To develop an end-to-end analytics pipeline for transforming unstructured online reviews into actionable business insights.
• To compare the effectiveness of various Machine Learning (ML) and Deep Learning (DL) models for sentiment classification and topic modeling in the smartphone market.

Main Methods:

• Utilized a dataset of ~68,000 Amazon reviews for ten smartphone brands.
• Employed class weighting to mitigate class-imbalanced data.
• Performed sentiment classification using ML (Decision Trees, Logistic Regression, SVM, Naive Bayes) and DL (CNN, RNN, LSTM) models.
• Extracted underlying themes using Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF).
• Assessed topic-level sentiment using the Valence Aware Dictionary and Sentiment Reasoner (VADER).
• Visualized results via an interactive Tableau dashboard.

Main Results:

• Convolutional Neural Network (CNN) achieved the highest sentiment classification accuracy (85.07%), performing well on positive and negative reviews but struggling with neutral ones.
• Non-Negative Matrix Factorization (NMF) topic modeling yielded more interpretable topics (Coherence Score=0.54) compared to Latent Dirichlet Allocation (LDA) (Coherence Score=0.41).
• Linked specific features like battery life and camera quality to customer sentiment.

Conclusions:

• The integrated pipeline, combining CNN for sentiment analysis and NMF for topic modeling, provides richer, business-ready insights than sentiment analysis alone.
• Interactive visualization enables practitioners to track sentiment trends, explore topics, and compare brand performance effectively.
• Acknowledged ethical considerations including potential bias and proposed future work on cross-domain generalization and fairness.