Alumni Interaction and Conversation Dynamics

Overview

This Natural Language Processing (NLP) project took a dive into chat engagement dynamics within the University of Ilorin’s Class of 2018 Statistics alumni group. By applying Latent Dirichlet Allocation (LDA) for topic modeling and network analysis, I was able to uncover communication patterns, topic distributions, and member interactions.

Dataset

The dataset was sourced from 'Certified Statistician' WhatsApp group consisting of 3,091 messages with the following breakdown:

• Average message length: 28.14 characters
• Media messages: 343
• Emojis used: 1,040
• Links shared: 33

Challenges

• Privacy and Ethical Considerations. Analyzing WhatsApp conversations involves accessing potentially sensitive personal data. To address this, I ensure the anonymization of alumni phone numbers and any other identifying details, safeguarding privacy while maintaining ethical standards throughout the analysis process.

• Informal Language. WhatsApp conversations often include slang, emojis, and informal language, which can complicate sentiment analysis.

• Data Noise. WhatsApp conversations contain excessive noise, such as irrelevant content, off-topic discussions, or incomplete sentences, which can complicate the sentiment analysis process. Cleaning the data and filtering out irrelevant content is essential to obtaining meaningful insights.

Methodology

1. Data Preprocessing

   • Cleaning and structuring raw chat data
   • Tokenization and stopword removal
   • Feature extraction (e.g., message types, emoji counts, link frequency)

2. Topic Modeling (LDA)

   • Optimized hyperparameters using Bayesian optimization
   • Extracted key discussion themes
   • Identified dominant topics and unique conversation threads

3. Network Analysis

   • Mapped participant interactions
   • Identified highly active members and community structure
   • Analyzed engagement levels and isolated nodes

---

Key Findings

• The conversation is predominantly faith-based, with topics such as Blessings & Devotion, Religious Messages, and Encouragement & Celebration.
• Topic 4 stands out as a distinct cluster, representing birthday celebrations, well-wishing, and encouragement messages.
• The most active participants, @STA14_1 and @STA14_6, form the core of the network, with @STA14_1 having the highest connectivity.
• Several participants exhibit minimal engagement, with messages that are either occasional replies or isolated posts.

Technologies Used

• Python (for data preprocessing, LDA, and network analysis)
• Gensim (for topic modeling)
• NetworkX (for social network analysis)
• Matplotlib, Seaborn, Plotly Express, and pyLDAvis (for data visualization)

Future Work

• Expanding the dataset to capture long-term engagement trends
• Applying sentiment analysis to understand emotional tones within topics
• Investigating time-based engagement patterns