PA164 Natural Language Learning (Strojové učení a přirozený jazyk)

Vážení,
online kurs (via ZOOM) se bude skládat z přednášky a projektu.
Předpokládá se znalost strojového učení v rozsahu IB031 Úvod do strojového učení a
základy počítačové lingvistiky.

Typicky, první hodina 9-10,00 bude věnována konzultacím k projektu.

10-10,15 bude interval pro zodpovězení miniodpovědníku (3 testové otázky, otevřen max. 5 minut)

10,15-11,50 bude online přednáška.

Diskusní fórum k organisaci kursu

Minitests 10 x 3 questions x 0.5 point = 15 p. minimum to pass: 6 p.

Project 50 p.

Final exam: test (Jan 11, 10,00) + oral exam if needed 25+10 p. min 15 p.

<50 F, <60 E, <70 D, <80 C, <90 B, >=90 A; zápočet: >= 45 p.

Extras: max. twice (i.e. max. 20 points)

Presentation on deep learning for NLP (a presentation 10-15 minutes + slides(tex/ppt/etc. + pdf) + report) max.2, 10 p. per talk per presentation

NEW: ML challenges in NLP, January 11.

Project (here) and at ISMU Learning Materials -> Project

Contents

• L1 (2 lessons)
  • 1. Organisation. Overview of machine learning
  • 2. Mining Text Data. Main tasks
  • 3. Project. Framework. Datasets. Tasks
• L2 Text representation and preprocessing
• L3-4 Text categorisation
• L5.1 Document sentiment analysis
• L5.2 Machine learning for disambiguation
  On Ambiguity
  Intro to disambiguation
  Penn Treebank tags
  Eric Brill Transformation-based tagger
  Memory Based Shallow Parser
  POS Tagging (State of the art)
  Flair and NAACL-HLT 2019 paper
• L6 Information Extraction
  Genic Interaction
  Genic Interaction Extraction Challenge
  Proceedings of LLL'05 Learning Language in Logic
  Identification of situations and actions in news articles (1st) (Rules learned) (paper IS-LTC'06) (c4c)
  
• L7 Deep learning for NLP
• L8 Consultations to the project
• L9.1 Outlier Analysis for Text
• L9.2 Keyness
  ACM Computing Classification System
  Keynesss and key words. In ISMU Study materials.
  Wordsmith tools, its keyness definition, Chi-squared test and Mike Scott page
• L10 Text summarization
• L11 Unsupervised methods. Document and term clustering
• L12 Jan 11: Final test. Projects. Future and challenges of ML for NLP Case study: PoS Tagging, ILP and Active Learning
  Future and challenges of Text mining and Natural language learning