Extracting keywords from texts
summary
Constructs a word co-occurrence network and ranks nodes by average edge weight rather than linguistic features, implemented in Python with NetworkX as SBKE using selectivity and a tunable generalized selectivity parameter. Tested on Croatian news, English Wikipedia and a Serbian–English parallel abstracts corpus; per-document Croatian F1 ≈25% rising to ≈34% for collection extraction, and corpus F1 ≈49.6% (English) and ≈46.7% (Serbian) when ignoring ~20% out‑of‑vocabulary keywords. Works without POS or language resources, yields high recall/low precision, and only extracts words that appear in the text.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:56Defining keyword extraction
- 2:34The big picture
- 4:07Constructing the network
- 6:05Network centrality measures
- 7:36The selectivity measure
- 8:50Experimental results
- 11:03Refining the method
- 12:18Example analysis
- 13:59Performance metrics
- 15:25Generalized selectivity
- 17:04Multilingual comparison
- 19:22Conclusion and summary
- 21:27Future work
- 23:51Key takeaway
- 25:17Q&A session
related talks
37mLexical, Vector & Hybrid Search with Elasticsearch
36mSAMENVATTR, or, How to Make a Text Summarization Tool for your Language of Choice
26mImplementing a Search Index
48mIntroduction to Natural Language Processing with Python
01h 41mGraph Intelligence: Enhance Reasoning and Retrieval Using Graph Analytics
35mHow does text become data?
58mAutomatic text summarization
22mMaking Natural Language Processing Robust to Sociolinguistic Variation