Topic modeling with BERTopic
summary
Applies BERTopic with Transformer embeddings plus UMAP and HDBSCAN to ~38,488 Moroccan tech tweets and produces a ranked list of candidate conference tracks. Pipeline surfaces about 5–7 coherent clusters: a logo design cluster (top cluster ~171 tweets), a North African dialect cluster (~166 tweets), a small JavaScript cluster (~26 tweets) and an FTX/crypto cluster (~14 tweets). Outputs labeled topics and example keywords, demonstrating a working suggestion engine that gives actionable track names and sample tweets.
AI-written summary · about
chapters (4)
- 0:00Introduction
- 10:04Topic modeling overview
- 23:00BERTopic implementation demo
- 31:18Conclusion and Q&A
related talks
24mBertTopic, From Free-Text feedbacks to Calls for Action
27mUnveiling Clustering in BERTopic Topic Modeling
13mThinking of Topic Modeling as Search
45mDecoding Topics: A Comparative Analysis of Python’s Leading Topic Modeling Libraries Using Climate C
35mLeveraging LLM for text augmentation in Topic Modeling
32mBERTopic to Accelerate Ukrainian Aid by the Red Cross
41mDesign and Implementation of a Topic Detection System
29mThinking of Topic Modeling as Search