poshlovesdata/ecommerce-sentiment-agents

Aspect-based multi-agent consumer analytics project for e-commerce reviews. Uses Jumia Nigeria mobile accessories as a case study, combining web scraping, rules-based NLP, LLM-powered aspect extraction, VADER comparison, and a Streamlit dashboard to reveal product-level sentiment insights beyond star ratings. # Aspect-Based Multi-Agent Consumer Analytics Research project for COSC878 - Web and Social Media Analytics. ## Title An Aspect-Based Multi-Agent Framework for Consumer Behavior Analytics: A Case Study of the Mobile Phone Accessories Market on Jumia Nigeria ## Project Goal This project collects product review data from Jumia Nigeria's mobile phone accessories market, processes the unstructured review text through a multi-agent NLP pipeline, and visualizes aspect-level consumer sentiment in a Streamlit dashboard. The implementation is organized to match the research methodology: 1. Data collection from public product review pages. 2. Text cleaning and product routing. 3. Aspect extraction from review text. 4. Aspect-specific sentiment classification. 5. Dashboard visualization and VPS deployment. ## Repository Structure ```text . |-- app/ | `-- streamlit_app.py # Phase 4 dashboard entrypoint |-- data/ | |-- processed/ # Cleaned/enriched review datasets | `-- raw/ # Raw scraped review datasets |-- deployment/ | |-- Dockerfile # Phase 5 container image | `-- docker-compose.yml # Phase 5 Traefik/VPS setup |-- notebooks/ # Optional exploration for the paper |-- reports/ | `-- figures/ # Exported plots for the research paper |-- scripts/ | |-- run_pipeline.py # Phase 3 batch processing CLI | `-- scrape_jumia.py # Phase 2 scraping CLI |-- src/ | `-- jumia_aspect_agents/ | |-- agents/ # Multi-agent NLP components | |-- analysis/ # Sentiment/aspect aggregation logic | |-- config.py # Environment-based settings | |-- data_collection/ # Jumia scraping modules | |-- models/ # Pydantic data contracts | `-- utils/ # Loguru, IO, retry helpers |-- tests/ # Unit and integration tests |-- .env.example …