A social-listening platform that scrapes Facebook and Instagram comments across multiple cosmetics brands, classifies them with a fine-tuned BERT model, and surfaces the results through a dashboard and exportable reports.
Business Problem
Marketing and customer-care teams across 8+ cosmetics brands had no systematic way to monitor high volumes of Facebook and Instagram comments, making it hard to spot delivery complaints, pricing questions or negative sentiment before they escalated.
Users
Marketing and customer-care teams monitoring brand reputation and customer questions across multiple social accounts.
Constraints
Solution
I built a pipeline that scrapes comments from each brand's Facebook and Instagram Graph API, normalizes them into a per-brand, per-platform PostgreSQL schema, and classifies each comment with a fine-tuned BERT sequence-classification model. The model's 11 output classes are mapped into a 9-category business taxonomy — FAQ, product availability, delivery issues, pricing, competition, engagement, tone and filter, plus positive/negative — surfaced through a dashboard with cross-brand analysis and exportable PDF reports.
Architecture
Your Contribution
Technology Stack
Results
Challenges & Trade-Offs
Cross-platform normalization vs. platform-specific nuance
Normalizing Facebook and Instagram comments into one per-brand schema keeps cross-brand analysis simple and consistent, trading away some platform-specific signal — comment threading, reaction types — that a platform-native pipeline could otherwise use.
Confidentiality
Enterprise project. Real brand names, comment content and account identifiers are not shown — this page describes the pipeline and taxonomy, not client data.