Projects

Starbucks Review Analysis

Data
NLP
Python

Data analysis project for sentiment prediction and location recommendation based on Starbucks reviews. Made to complete the qualification case to teach Natural Language Processing as a lab assistant.

View Source Code

Starbucks Review Analysis Cover

Starbucks Review Analysis is a data pipeline and analysis project that predicts the sentiment of a Starbucks review and gives Starbucks location recommendations, showcasing understanding and skills in the AI subfield of Natural Language Processing (NLP).

This project was completed as a qualification case (with three parts) to teach Natural Language Processing (NLP) as a laboratory assistant. I chose a dataset of Starbucks customer reviews because I wanted to learn about the process of sentiment analysis for reviews, and (incidentally) I am an avid coffee drinker. There are three notebooks altogether, which demonstrate a foundational data pipeline, from raw data ingestion and cleaning to feature extraction and modeling. From this project, I developed skills in data manipulation, text processing, and building predictive models, which are fundamental to a data engineering workflow.

Tech Stack: Python with libraries Pandas, NLTK, Scikit-learn, Matplotlib, spaCy, pickle (requirements.txt for setting up virtual environment)

Dataset

The data for this project was sourced from a Kaggle dataset created by web scraping the ConsumerAffairs website. It provides a comprehensive collection of consumer reviews and ratings for Starbucks, offering valuable insights into customer sentiment.

The dataset contains the following columns:

  • Name: The reviewer’s name.
  • Location: The location or city associated with the reviewer.
  • Date: The date the review was posted.
  • Rating: The star rating given by the reviewer (1 to 5).
  • Review: The full text content of the review.
  • Image Links: Links to any images included with the reviews.

Key Features

Case 1: Starbucks Reviews Sentiment Analysis with Naive Bayes

Sentiment Analysis Prediction

  • Data Ingestion & Cleansing → Loads the .csv file into a Pandas DataFrame, drops rows with missing values, and converts date strings to a standard datetime format for analysis.
  • Exploratory Data Analysis (EDA) → Visualizes key trends to find insights such as the distribution of ratings, the states with the most reviews, and the busiest months.
  • Text Preprocessing → Cleans raw text by removing special characters, tokenizing words, removing stop words, and performing lemmatization.
  • Frequency Distribution → Finds the most common words in positive and negative reviews to understand sentiment drivers.
  • Model Building & Persistence → Trains a Complement Naive Bayes (ComplementNB) classifier for sentiment analysis and saves the model using Pickle. ComplementNB is excellent for imbalanced datasets like these.

Case 2: Starbucks Location Recommendations

Location Recommendation

  • N-Gram Modeling → Implements an n-gram model to analyze word sequences and calculate text similarity.
  • TF-IDF for Similarity → Uses TF-IDF to vectorize text data and calculate cosine similarity to recommend similar Starbucks locations based on a user’s review.
  • Named Entity Recognition (NER) → Uses spaCy to identify and extract entities like locations, organizations, and people from review text.

Case 3: Grammar Parsing using Natural Language ToolKit

Grammar Parsing

  • Context-Free Grammar (CFG) → Defines a simple grammar using NLTK to demonstrate how to parse sentences and understand their syntactic structure.
  • Syntactic Analysis → The parser builds parse trees for sample sentences, visually representing their grammatical structure.