Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Ensemble Topic Modelling

Sponsored · Your Podcast. Everywhere. Effortlessly. Share. Educate. Inspire. Entertain. You do you. We'll handle the rest. →
Avatar for Leland McInnes Leland McInnes
July 12, 2019
Research
1
480

Ensemble Topic Modelling

A short lightning talk on ensemble topic modelling with pLSA using the enstop package.

Avatar for Leland McInnes

Leland McInnes

July 12, 2019
Tweet

More Decks by Leland McInnes

See All by Leland McInnes
PyNNDescent: Fast Approximate Nearest Neighbors with Numba
Avatar for Leland McInnes lmcinnes
0
1k
Word and Document Embeddings
Avatar for Leland McInnes lmcinnes
0
160
Topological Data Analysis
Avatar for Leland McInnes lmcinnes
1
340
Learning Topology: topological methods for unsupervised learning
Avatar for Leland McInnes lmcinnes
2
3.6k
A Guide to Dimension Reduction
Avatar for Leland McInnes lmcinnes
3
1.4k
UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
Avatar for Leland McInnes lmcinnes
2
2.7k

Other Decks in Research

See All in Research
令和最新技術で伝統掲示板を再構築: HonoX で作る型安全なスレッドフロート型掲示板 / かろっく@calloc134 - Hono Conference 2025
Avatar for calloc134 calloc134
0
560
2025-11-21-DA-10th-satellite
Avatar for Yuka Egusa yegusa
0
130
SREはサイバネティクスの夢をみるか？ / Do SREs Dream of Cybernetics?
Avatar for Yuuki Tsubouchi (yuuk1) yuukit
3
410
LLM-jp-3 and beyond: Training Large Language Models
Avatar for Yusuke Oda odashi
1
770
A History of Approximate Nearest Neighbor Search from an Applications Perspective
Avatar for Yusuke Matsui matsui_528
1
180
"主観で終わらせない"定性データ活用 ― プロダクトディスカバリーを加速させるインサイトマネジメント / Utilizing qualitative data that "doesn't end with subjectivity" - Insight management that accelerates product discovery
Avatar for 株式会社カミナシ kaminashi
15
22k
その推薦システムの評価指標、ユーザーの感覚とズレてるかも
Avatar for kuri8ive kuri8ive
1
330
Combining Deep Learning and Street View Imagery to Map Smallholder Crop Types
Avatar for SatAI.challenge satai
3
630
情報技術の社会実装に向けた応用と課題：ニュースメディアの事例から / appmech-jsce 2025
Avatar for Shotaro Ishihara upura
0
330
svc-hook: hooking system calls on ARM64 by binary rewriting
Avatar for Akira Moroo retrage
1
130
生成AI による論文執筆サポート・ワークショップ 論文執筆・推敲編 / Generative AI-Assisted Paper Writing Support Workshop: Drafting and Revision Edition
Avatar for Kenji Saito ks91
PRO
0
130
生成的情報検索時代におけるAI利用と認知バイアス
Avatar for Y. Yamamoto trycycle
PRO
0
340

Featured

See All Featured
Everyday Curiosity
Avatar for Cassini Nazir cassininazir
0
140
Hiding What from Whom? A Critical Review of the History of Programming languages for Music
Avatar for Tomoya Matsuura tomoyanonymous
2
460
Lightning talk: Run Django tests with GitHub Actions
Avatar for Sarah Abderemane sabderemane
0
130
A designer walks into a library…
Avatar for Paul Jervis Heath pauljervisheath
210
24k
Leadership Guide Workshop - DevTernity 2021
Avatar for David Neal reverentgeek
1
220
Navigating Team Friction
Avatar for Lara Hogan lara
192
16k
Crafting Experiences
Avatar for Bethany Jepchumba bethany
1
65
Kristin Tynski - Automating Marketing Tasks With AI
Avatar for Tech SEO Connect techseoconnect
PRO
0
170
How to build a perfect <img>
Avatar for Jono Alderson jonoalderson
1
5.2k
The #1 spot is gone: here's how to win anyway
Avatar for Tamara Novitovic tamaranovitovic
2
960
GitHub's CSS Performance
Avatar for Jon Rohan jonrohan
1032
470k
Why Your Marketing Sucks and What You Can Do About It - Sophie Logan
Avatar for Sophie Logan marketingsoph
0
85

Transcript

  1. Ensemble Topic Modelling Leland McInnes

  2. Model a corpus of documents in terms of underlying “topics”

  3. Topic Modelling as Matrix Factorization

  4. None
  5. None
  6. None
  7. None

  8. LDA and pLSA are probabilistic matrix factorization methods

  9. (Ensembles of) pLSA

  10. Performance?

  11. None

  12. Quality?

  13. None

  14. Instability?

  15. These are hard optimization problems

  16. Topics vary from one run to another

  17. What are the stable topics? Inspired by https://github.com/RaRe-Technologies/gensim/pull/2282

  18. None

  19. Each cluster represents a stable topic

  20. None

  21. • Greater stability • Determines number of topics automatically •

    Embarrassingly parallel computation

  22. Implementation

  23. sklearn API

  24. None

  25. https://github.com/lmcinnes/enstop

  26. pip install enstop