Objective / Business Problem
A leading retail company aimed to uncover the key factors that drive customers to recommend their products by analyzing customer reviews. The goal was to predict product recommendation status directly from review text, reducing dependence on explicit recommendation feedback.
Approach / Methodology:
-
Data Collection: Scraped product reviews, associated satisfaction scores, and whether customers recommended the product.
-
Text Preprocessing:
- Tokenization: Reviews were split into sentences and words; words converted to lowercase; punctuation, stop words, and words with fewer than three characters were removed.
- Lemmatization: Words were normalized to their base forms, including converting third-person verbs to first person and standardizing tense.
- Stemming Words were reduced to root forms to unify similar terms.
- Feature Extraction: Applied TF-IDF to transform cleaned text into numerical feature vectors representing term importance across the corpus.
- Feature Selection: Used methods like LASSO regression and filter-based techniques to identify key predictive features from the TF-IDF matrix.
- Modeling: Constructed a supervised machine learning pipeline applying multiple algorithms on shortlisted features, selecting the best-performing model based on accuracy metrics.
Outcomes and Impact:
-
Developed a robust classification algorithm that predicts whether a customer will recommend a product solely from their review text.
-
Eliminated the operational need to explicitly collect recommendation scores from customers, streamlining feedback analysis.
-
Provided actionable insights into the language and sentiments associated with product endorsement, guiding marketing and product improvement strategies.
-
Enabled proactive targeting of customers with low likelihood to recommend, allowing tailored interventions to boost satisfaction and advocacy
Business Value:
This predictive text classification system enhances customer insight extraction, improves operational efficiency by reducing data collection burden, and supports evidence-based decision-making to increase product recommendation rates and customer loyalty.
