This project presents a comprehensive case study on Women's E-Commerce Clothing Reviews, focusing on two key areas: Sentiment Analysis and Predictive Modeling. The primary objective was to extract actionable insights from unstructured text data to understand customer sentiment and predict how a product might be rated based on the language of the review.
Conducted a comparative sentiment analysis using multiple popular NLP libraries:
-
spaCy: Leveraged for its powerful linguistic features and text processing capabilities to prepare the data for analysis.
-
TextBlob: Employed for its straightforward and accessible API to get polarity and subjectivity scores.
-
NLTK's VADER (Valence Aware Dictionary and sEntiment Reasoner): Utilized for its effectiveness on social media and short, informal text, which is common in reviews.
-
The Positive WordCloud visually confirmed that attributes like fit, comfort, color, and quality (using words like 'love', 'perfect', 'great', 'soft', 'beautiful') are key drivers of high ratings.
-
In stark contrast, the Negative WordCloud immediately highlighted significant customer pain points, with words like small, tight, cheap, return, and disappointed being most prominent.
-
Text Preprocessing & Feature Engineering: Cleaned and prepared the text data. Utilized TF-IDF Vectorizer to convert the corpus of reviews into a meaningful matrix of numerical features, giving weight to words that are more important to a specific review.
-
Data Visualization: Generated WordClouds for different sentiment categories to visually identify the most frequent and prominent words associated with positive and negative feedback
-
Model Building: Implemented a Logistic Regression model, a robust and interpretable classification algorithm, to predict the sentiment or star rating of a review. The model was trained on the TF-IDF features to learn the patterns in the language that correlate with customer satisfaction.
- Language: Python
- Environment: Jupyter Notebook
- Pandas: Used for loading, cleaning, and structuring the e-commerce review dataset.
- spaCy: Leveraged for efficient text preprocessing, including tokenization and lemmatization.
- NLTK (VADER): Implemented for its powerful sentiment analysis on short, informal review texts.
- TextBlob: Used for comparative sentiment scoring to validate results.
- Scikit-learn: Employed for:
TfidfVectorizer: To convert text data into numerical features.LogisticRegression: To build and train the predictive model.
- WordCloud: To create insightful visualizations of the most frequent positive and negative words.
- Matplotlib & Seaborn: For generating plots and charts during exploratory data analysis (EDA).
That's all folks