BERTopic to Accelerate Ukrainian Aid by the Red Cross
summary
BERTopic, an LLM-based topic modeling method, is explained at a high level and evaluated with three metrics (including a GPT-generated test set). The talk shows how the Netherlands Red Cross used BERTopic on Telegram social-media posts to quickly extract topics and insights about Ukrainian refugees’ needs so aid could be targeted more effectively.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:43Topic modeling for aid
- 3:13Red Cross social media
- 5:20BERTopic processing
- 7:48Clustering in BERTopic
- 11:02BERTopic settings
- 12:50BERTopic speed
- 13:11Implementing BERTopic
- 14:33Visualizing topics
- 16:45Summarizing topics
- 18:12Evaluating BERTopic
- 21:23Topic accuracy
- 23:19Sentence transformers
- 24:03BERTopic impact
- 25:52Future applications
- 26:15Q&A
related talks
34mTopic modeling with BERTopic
24mBertTopic, From Free-Text feedbacks to Calls for Action
43mDiscovering the BERT family of models
57mDiscover semantic topics in a collection of documents
57mNatural Language Processing 2020 Discussion
41mRepresentations from natural language data: successes and challenges
20mBI Toolkit Powered by Natural Language Processing
43mAdvanced Prompt Engineering and Beyond