Automatic text summarization
summary
Covers extractive vs abstractive approaches with concrete methods and outcomes: positional heuristics find the topic sentence 85% first sentence, 7% last; Luhn style scoring uses concentrated term frequency; centroid and graph/PageRank methods select central sentences; max marginal relevance reduces redundancy. Abstractive models use sequence-to-sequence (2014) encoder–decoder networks and still struggle with world knowledge. Evaluation uses precision/recall, utility and ROUGE recall-based n-gram overlap and pyramid weighting for multi-document summaries. Tools: Python and TensorFlow.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:37Automatic text summarization
- 6:25Extractive vs. Abstractive Summary
- 8:33Extractive Summaries-Lack of balance
- 9:36Extractive Summaries-Lack of cohesion
- 13:02Positional method
- 14:41Luhn's method
- 17:43Edmundson's method
- 20:58FRUMP - Demonstration script
- 22:10Classification
- 25:56Maximal marginal relevance
- 34:48Sequence to sequence
- 38:34What makes a good summary?
- 40:21Types of evaluation methods
- 41:57Precision and Recall
- 45:50Utility
- 51:47Pyramid method
related talks
22mEmpowering journalism-approaches to automatic text summarization
36mSAMENVATTR, or, How to Make a Text Summarization Tool for your Language of Choice
19mPhrazIt : Tool for automatic text summarization
01h 31mNatural Language Understanding: Foundations and State-of-the-Art
26mUnsupervised learning for news summarisation
06mHierarchical Summarization for Longform Spoken Dialog
28mSemantic SentenceRank to Facilitate Reading....
31mSAMEntics : Tools for paraphrase detection