Fragrance Industry Research & Data Integration

A well-known UK university wanted a deeper, data-driven view of the global fragrance industry — how scents are made, marketed, and perceived. We partnered with its computational researchers on the full pipeline: scraping and integrating multi-source data, building bespoke NLP models, running statistical and network analysis, and delivering results through interactive dashboards and academic conferences.

The problem

The signal the researchers needed was scattered across product catalogs, consumer reviews, user forums, and social media — each with its own naming, structure, and noise. Before any analysis could happen, that data had to be collected lawfully, linked accurately, and stored at scale.

What we built

Data acquisition and governance

• Automated scraping with requests and Selenium pulled real-time product details, consumer reviews, and social media mentions.
• Compliance built in — data practices aligned with the UK's Data Protection Act and GDPR, with secure handling of sensitive user-generated content.

Data linkage and integration

• Fuzzy matching algorithms merged disparate product catalogs and consumer databases at 98.6% accuracy, unifying thousands of records under consistent naming and classification.
• PostgreSQL and data-lake storage handle growing data volumes without degrading performance.

NLP and sentiment analysis

• Transformer embeddings (BERT, DeBERTa v2) classify fragrance notes and gauge consumer sentiment toward new and niche products.
• Llama 2 for JSON extraction parses complex text blurbs from user forums and professional reviews into structured JSON.
• K-means clustering groups similar feedback, and BART-CNN generates automated summaries that turn dense text into readable findings.

Statistical and network analysis

• pandas, NumPy, Statsmodels, and SciPy power everything from hypothesis testing and regression to time-series analysis.
• Custom similarity metrics and novelty scores assess innovation in fragrance formulas and identify top performers by unique scent composition.
• NetworkX maps key influencers, collaboration patterns between perfumers, and the dynamics of consumer communities.

Visualization and reporting

Interactive dashboards built with Plotly and Tableau let researchers explore trends, notes, and market behaviors in real time. Findings were presented at academic conferences, with non-technical reports and executive dashboards for broader audiences.

Results

• A unified view of the fragrance ecosystem. Combining product catalogs, user reviews, and social media chatter gave researchers access the field had not had before.
• Reliable foundations for inference. The 98.6% integration accuracy cut discrepancies and raised confidence in every downstream statistical model.
• Academic and commercial value. The work informed theoretical models of consumer behavior and innovation while yielding practical intelligence for fragrance companies.
• Results everyone could use. Clear reports and intuitive dashboards meant technical and non-technical stakeholders alike could act on the findings.