💬 ChatMeter : AI Competitive Intelligence Solution 💡

📝 Description

ChatMeter is an AI-powered solution designed to analyze and understand the competitive landscape of telecommunications operators in the Portuguese 🇵🇹 market. By monitoring social media, it provides businesses with crucial insights into customer sentiment, emerging topics, and competitor strategies, acting as a real-time "thermometer" for the market.

🎓 Project Context

This project was developed for the Projeto Final Aplicado a Ciência de Dados (Final Applied Project in Data Science I & II) courses. This semester-long capstone project is a central part of the 3rd and final year of the Licenciatura em Ciência de Dados (Bachelor Degree in Data Science) at ISCTE-IUL, completed in the 2023/2024 academic year. With this project, we aim to apply the knowledge and skills acquired throughout the course to a real-world problem, leveraging data science techniques to provide actionable insights and solutions for a real business challenge.

✨ Objective

The primary objective of ChatMeter is to develop a "Sentiment and Topic Thermometer" for Portuguese telecommunications operators. This tool enables stakeholders to:

• Identify user needs, preferences, and behaviors on social media.
• Discover new opportunities, trends, and challenges within the sector.
• Support data-driven, informed decision-making in a competitive market.

🏗️ Project Workflow (CRISP-DM)

This project followed the CRISP-DM methodology for data mining. Here's a breakdown of the key activities in each phase:

1. Business Understanding: 💡
   • Defined project goals: Analyze social media data to understand the competitive landscape of telecom operators.
   • Identified key players: Vodafone 🔴, MEO 🔵, NOS ⚫, and DIGI 🟡.

2. Data Understanding: 🔍
   • Collected data from Facebook pages of telecom operators using web scraping.
   • Explored the structure and format of the collected data (posts, comments, user information).
   • Assessed data quality and identified potential issues (missing values, duplicates).

3. Data Preparation: 🛠️
   • Data Cleaning: Removed duplicate entries, handled missing values, and corrected inconsistencies in the data.
   • Text Preprocessing: Converted text to lowercase, removed URLs, punctuation, stopwords, and applied stemming and lemmatization.
   • Feature Engineering: Created new features, such as a binary variable to indicate if the text mentions a specific operator.

4. Modeling: 🤖

   • Sentiment Analysis: Applied pre-trained transformer models (TXRBSF and mDeBERTa) to classify the sentiment of the text ( positive / neutral / negative ).
   
   
   • Topic Analysis: Utilized Zero-Shot Classification with the mDeBERTa model to identify the main topics discussed in the text.

5. Evaluation: ✅
   • Create a new matrix to combine the sentiment and topic analysis results.
   • Tested for significance using the Chi-Squared Test.

6. Deployment: 🚀

   • Developed an interactive dashboard using Streamlit to visualize the results and insights.

---

🗺️ Workflow

The project's end-to-end workflow is summarized below:

---

📊 Interactive Dashboard

Access the live interactive dashboard here: ChatMeter - Streamlit App

👥 Team Members (Group 5)

• André Silvestre (Nº104532)
• Eliane Gabriel (Nº103303)
• Maria João Lourenço (Nº104716)
• Maria Margarida Pereira (Nº105877)
• Umeima Adam Mahomed (Nº99239)

🇵🇹 Note

This project was developed in PT-🇵🇹.