Analyzing Customer Experience: Multi-Platform Review Analysis for Food Delivery Apps

Welcome

This project presents a comprehensive analysis of customer experiences across popular food delivery platforms, using user-generated reviews from multiple sources.

We explore natural language processing (NLP) techniques to understand what customers say about key aspects of service such as delivery time, app usability, food quality, and customer support. By leveraging pretrained models and modern text classification methods, we extract meaningful insights to support product improvement and competitive analysis.

Summary of Findings

This project investigates customer experiences with major food delivery platforms — Uber Eats, DoorDash, and Grubhub — by analyzing unstructured reviews from Reddit, the App Store, and Google Play. The analysis reveals that disappointment is the most commonly expressed emotion, particularly in App Store reviews for Grubhub and Uber Eats.

Negative sentiment is largely tied to late or missing orders, delivery delays, app glitches, and unresponsive customer service. DoorDash also sees notable complaints about pricing issues.

Topic modeling uncovered four dominant themes across reviews:
1. Service quality problems
2. Delivery inefficiencies
3. App-related frustrations
4. A smaller portion of positive feedback

Based on these insights, we recommend targeted improvements in customer support, logistics, app usability, and pricing to boost user satisfaction and retention.

Objectives

  • Analyze and visualize customer sentiment across multiple food delivery apps
  • Apply zero-shot classification to identify review topics without labeled data
  • Use emotion and sentiment analysis to detect emotional tones in customer feedback
  • Perform exploratory data analysis (EDA) to uncover review patterns and trends
  • Apply topic modeling to identify hidden themes in large text corpora

Analysis Modules

Explore the full analysis by navigating through the sections below:

  • 🔍 Zero-Shot Classification
    Use transformer models to classify reviews into topics without predefined training labels.

  • 😠 Emotion Analysis
    Detect the emotional undertone (joy, anger, fear, etc.) in customer reviews.

  • 🙂 Sentiment Analysis
    Classify review polarity (positive, negative, neutral) using pretrained sentiment models.

  • 📊 Exploratory Data Analysis
    Understand review distributions, rating patterns, and platform-level trends.

  • 🧠 Topic Modeling
    Discover key themes in large text datasets using unsupervised learning (LDA/BERT).

Group 19

This project was completed by Group 19 as part of MSDS 597 – Data Wrangling at Rutgers University, Spring 2025.