Business Objective
Our client is a European market insights firm in the Food & Beverage industry.
Their core offering is a one-of-a-kind platform which tracks global product launches and provides insights on product launches and their attributes such as ingredients, flavors, etc. They had two objectives – detect popular themes from product descriptions of packaged F&B products, and corroborate this using stated preferences on social media.
Challenges
- Posts on social media, especially tweets, tend to be short and normal rules of grammar often do not apply
- Eliminating redundancy by identifying and eliminating synonyms in addition to standard text parsing and data cleansing
- 2+ million unique products across 125+ F&B categories, 10+ years, and 75+countries.
Solution Methodology
- Processed text of the product labels and claims. The activity included parsing, cleansing, lemmatizing, and POS tagging
- Identified and acquired relevant tweets through the Twitter API. Parsed and cleansed the text to prepare it for modeling
- Identified topics – stable and transient – using variants of Latent Dirichlet Allocation models
- Used Bayesian classifiers to tag new tweets into existing topics, if appropriate
- Used fuzzy matching to map topics between processed documents and key F&Bthemes identified from tweets
- Developed an insights dashboard showing key trends for each F&B category, by geography.
Business Impact
- An automated platform that eliminates the need for manual research and subjective interpretation while detecting popular themes in a category.
- Ability to quickly generate customized, ready-to-consume insights for any audience.
