Decoding Topics: A Comparative Analysis of Python’s Leading Topic Modeling Libraries Using Climate C

on YouTube ↗45m

" summary

Compares LDA, Top2vec, and BERTopic on 14,000 climate policy text blocks and four industry tasks: similarity, classification, real-time insertion, and dynamic topic tracking. Finds LDA very fast but fails at similarity; Top2vec yields best similarity and stores document embeddings for quick cosine lookups; BERTopic gives superior visualizations and supports dynamic topic modeling but creates many outliers and needs extra tuning. Quantitative results: ~70 automatic topics from the embedding methods and up to 92% classifier accuracy using topic labels.

AI-written summary · about

" chapters (31)

1. 0:00Introduction and speaker background
2. 1:31What is topic modeling and why it matters
3. 4:21Why this talk and industry inspiration
4. 5:44Overview of topic modeling approaches
5. 7:53Use cases for topic modeling in industry
6. 11:25Data set selection and preparation
7. 13:10Implementation details for LDA
8. 13:52Implementation details for Top2Vec
9. 14:55Implementation details for BERTopic
10. 16:46Evaluation of libraries and models
11. 18:31Results for use case one finding similar documents
12. 20:37Results for use case two document classification
13. 22:44Results for use case three real-time integration
14. 25:06Results for use case four dynamic topic modeling
15. 27:35Comparison and personal recommendations
16. 29:45Q&A session introduction
17. 30:07Handling short documents like tweets
18. 32:37Classification task inputs and methods
19. 33:18Handling emerging or unseen topics
20. 34:02Using BERTopic embeddings for classification
21. 35:07Measuring topic distinguishability
22. 36:11Dynamic topic modeling and disappearing topics
23. 37:15Manual evaluation of topics with human annotators
24. 38:20Topic modeling with image or video data
25. 38:42Using n-grams in topic modeling
26. 39:25Can LLMs outperform topic modeling methods
27. 40:30Topic modeling on multilingual data
28. 41:32Merging smaller topics in LDA
29. 42:17Handling documents with multiple topics
30. 43:40Encoder models behind BERTopic
31. 44:01Topic modeling for unordered unrelated terms

" related talks

• 13mThinking of Topic Modeling as SearchKas Stohr · PyBay 2024
• 27mUnveiling Clustering in BERTopic Topic ModelingAbhiram Ravikumar, Jaspal Singh · Conf42 Machine Learning 2023
• 34mTopic modeling with BERTopicMounir Boulwafa · BlaBlaConf 2022
• 01h 01mWorkshop: Intro to topic modeling using PythonRussell Martin · Big Data & AI Toronto 2020
• 45mEvaluating Topic ModelsMatti Lyra · PyData Berlin 2017
• 57mDiscover semantic topics in a collection of documentsGabriela Zarzar Gandler · PyCon SE 2022
• 01h 26mBagging to BERT- A tour of applied NLPBenjamin Batorsky · PyData NYC 2022
• 35mLeveraging LLM for text augmentation in Topic ModelingAriel Gamiño · PyData Global 2023