John Bellanger - Sentiments Analysis: Back Pain

Transforming the Voice of the Customer into High-Performance Advertising

1. Context & Business Problem: How to Turn Listening into Conversion?

The objective of this project was to address a fundamental business challenge: how to design a hyper-effective advertising campaign for a back pain solution. Rather than relying on marketing intuition, my premise was that a rigorously data-driven approach would allow me to build a message with surgical precision.

The real question wasn't "what should we tell our customers?" but rather "what are our customers already telling each other?" To answer this, I decided to analyze the most authentic source available: the spontaneous conversations of hundreds of people on public forums, specifically Reddit.

2. My Data-Driven Methodology: An End-to-End Mastered Process

To transform raw discussions into a quantifiable marketing strategy, I deployed a complete processing pipeline within the R programming environment, a choice driven by its statistical power and high-quality visualization capabilities.

Data Source: Public discussions about back pain, scraped from Reddit using the Apify tool. This proactive approach allowed me to build a unique and unbiased dataset.
Analysis Tools:
1. R: For the entire process, from data cleaning to semantic analysis and visualization creation.
2. Google Slides: For the final communication of findings, in order to build a clear and direct narrative.
Data Collection & Cleaning: Raw text from the internet is inherently "dirty." I therefore applied a rigorous cleaning script in R, validating each step iteratively with the print() function to ensure data integrity. This process included:
1. tolower(): Standardizing case so that "Pain," "pain," and "PAIN" were treated as a single concept.
2. gsub("[[:punct:]]", "") & gsub("[[:digit:]]", ""): Removing punctuation and numbers, which are noise for semantic analysis.
3. removeWords(stopwords("english")): Removing common words (stopwords like "the," "a," "is") to retain only meaningful terms.
Analysis Technique: I used Term Frequency analysis to identify the most recurring themes. This is the most direct method for quantifying the main concerns of users.

3. Analysis & Insights: Moving from "What" to "Why"

Once the data was ready, I could begin to make it talk. My analysis unfolded in two phases.

A. Descriptive Analysis: Identifying the "What"

The first step was to quantify the topics of discussion. What was the number one concern for people suffering from back pain?

My Method: An In-Depth Analysis to Reveal the Unseen

To transform this raw text into a clear strategy, I followed a rigorous process using the R programming language, known for its statistical power.

Text Cleaning: The Foundation of a Sound Analysis. Text from the internet is "dirty." I therefore prepared the data using essential functions:
- tolower(): To standardize the text to lowercase, ensuring that "Pain," "pain," and "PAIN" are treated as the same word.
- gsub("[[:punct:]]", "") and gsub("[[:digit:]]", ""): To remove punctuation and numbers, which are "noise" for sentiment analysis.
- removeWords(stopwords("english")): To remove common stop words (e.g., "the," "is," "a") that lack semantic weight, allowing me to focus on meaningful words.
Sentiment and Emotion Analysis: Quantifying Opinion. I used the R library sentimentr, a specialized tool for this type of analysis.
- sentiment_by(): This function allowed me to assign a sentiment score (positive, negative, neutral) to each comment, giving me an overview of the general tone of the discussions.
- emotion_by(): Even more powerfully, this function detected the presence of 8 fundamental emotions (joy, sadness, anger, fear, etc.). It was this function that enabled me to understand the emotional charge behind the words.

The 'Aha Moment': Beyond the Pain, What's the Real Story?

The overall analysis immediately revealed a critical insight. While general sentiment is mixed, the emotional palette is dominated by negativity, fear, and sadness.

Key Observation: The problem of back pain is not just a physical issue. It is, first and foremost, an emotional suffering that profoundly impacts daily life.

Distribution of Emotions in Comments Sadness and fear are the most frequently expressed emotions by the audience.

Analysis with R

This initial analysis confirmed the obvious themes. But to create a powerful message, I needed to dig deeper and understand the relationship between them.

B. Diagnostic Analysis: Understanding the "Why"

The true insight isn't found in the words themselves, but in the context in which they are used. Why is "pain" such a central problem? Is it just a physical sensation?

To find out, I conducted a bigram analysis (most frequent pairs of words). My hypothesis was that physical pain had direct consequences on emotional well-being and daily life.

My Hypotheses: From Observation to Strategy

This initial discovery guided my strategy. I formulated three hypotheses to structure my analysis and turn it into concrete actions.

Hypothesis 1: Is the problem truly more emotional than physical?

Why this hypothesis? To confirm that the core of the ad message needed to be empathy.
Verification: I isolated the "Negative" comments and analyzed the emotions they contained. The result was unequivocal.
Conclusion: Hypothesis validated. The discussion around pain is intrinsically linked to sadness, anger, and fear. Our advertising must speak to these emotions.

What are the hidden emotions behind back pain? Within negative comments, sadness, fear, and anger are overrepresented, confirming the customers' emotional distress.

Hypothesis 2: Are competing solutions a source of frustration?

Why this hypothesis? To determine if we should adopt a head-on attack strategy against the competition.
Verification: I analyzed the sentiment of comments that mentioned competitors (medications, doctors, etc.).
Conclusion: Hypothesis invalidated. The dominant sentiment is "Neutral." Customers discuss these solutions factually. Attacking the competition would therefore be an ineffective strategy.

Sentiment Towards Competing Solutions Neutrality dominates, indicating that an aggressive advertising strategy against competitors would be ill-advised.

Hypothesis 3: What words embody the ideal solution?

Why this hypothesis? Since we cannot define ourselves against the competition, we must define ourselves for what customers desire.
Verification: I created a word cloud from the "Positive" comments.
Conclusion: Hypothesis validated. The word cloud reveals the concrete benefits and the exact words customers use to describe their relief: "relief," "walk," "sleep," "finally." These are the words for our future ad.

The Words of Relief This word cloud provides the key arguments for our future advertising promise.

My Action Plan: Quantified and Predictive Recommendations

Based on these analyses, I have formulated a 3-stage action plan, and I have measured the potential impact of each stage with statistical models.

1. Immediate Recommendation: Launch an Empathetic Ad Campaign

What: Create an ad that leverages the emotional distress (H1) and the benefit-oriented words (H3).
Predictive Impact: A T-test, used to statistically compare two versions, showed that this new approach would generate a CTR of 2.08% versus 1.22% for the old one. This is a predicted increase of 70%, with 99.7% statistical confidence.

Is Empathy More Clickable? (A Predicted +70% CTR) The T-test confirms that the empathetic message significantly outperforms the generic approach.

2. Optimization Recommendation: Develop Reassurance Content

What: Create articles and testimonials focused on concrete benefits.
Predictive Impact: A Linear Regression model, which measures cause-and-effect relationships, predicts this content will increase the engagement rate by +2.5 percentage points. The near-zero p-value (2.66e-08) guarantees the reliability of this prediction.

Can Reassurance Content Truly Double Our Engagement? Our model predicts a significant increase in engagement thanks to a targeted content strategy.

Prediction with R

3. Preventive Recommendation: Build a "Champions" Community

What: Foster loyalty in our best customers by creating a support community.
Predictive Impact: A Clustering (k-means) algorithm, which segments the customer base, identified a group of "Champions" (10-20 purchases/year, $55 - $80 average order value). Targeting this segment will increase retention and Customer Lifetime Value (CLV).

Who Are Our 'Champions'? Clustering reveals a high-value customer segment, a priority target for our loyalty initiatives.

5. Technical Skills Demonstrated

Programming (R): Mastery of the complete data value chain: scraping (via Apify), complex text data cleaning, semantic analysis, and creating high-quality visualizations with ggplot2.
Statistical Analysis: Application of frequency and contextual analysis (bigrams) to transform unstructured qualitative data into quantifiable and actionable insights.
Strategic Vision & Communication: Ability to translate a data insight into a clear business recommendation. Deliberate choice not to create a dashboard for a one-time analysis, instead favoring a linear presentation on Google Slides for controlled and persuasive storytelling.

Presentation

Conclusion
This analysis demonstrates that data science, when applied to marketing, allows for the creation of an advertising strategy that is not only creative but also scientifically validated. By listening to the voice of the customer, we have discovered the shortest path to performance: talk less about our product, and more about the life they'll get back.
I am available to discuss the implementation of this action plan.

Let's become partner

Contact

Mail

Github

My Method: An In-Depth Analysis to Reveal the Unseen

The 'Aha Moment': Beyond the Pain, What's the Real Story?

My Hypotheses: From Observation to Strategy

My Action Plan: Quantified and Predictive Recommendations

Conclusion

Contact