Evaluation

• 20 p. poster

• 50 p. project

• 30 p. final exam (obligatory, must obtain at least 15 p.)

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

A Python notebook supporting the lecture part on typical NLP pipelines:

Additional readings:

QUESTIONS AND TASKS:

• Natural language (pre)processing techniques and their relevance for building machine learning models applicable to text

• Bag of words representation of text - pros and cons

QUESTIONS AND TASKS:

• Text representations. When the ordering of words (does not) matter-s. Mining web
  

• Main text mining tasks

• ML for disambiguation

• CNN for NLP

MATERIALS FOR THE LABS IN WEEK 02 (all necessary details in the notebook):

Slidy k prednasce (veskere reference apod. inline):

READINGS

QUESTIONS AND TASKS:

• Distributional hypothesis - historical context, linguistic motivations and practical implementations

• Distributional vs. formal semantics

• Latent semantic analysis - basic principles, pros and cons

• Word embedings - comparison of selected popular approaches

• Basic principles of language models and their training process

MATERIALS FOR THE LABS IN WEEK 04 (all necessary details in the notebook):

NN for NLP. Convolutional NN

Data augmentation

Tracking Progress in Natural Language Processing

Recurrent NN

Machine translation

Text clustering

QUESTIONS AND TASKS:

• Techniques for data augmentation

• NN for machine translation

• Text clustering

READINGS

Recurrent NN

Outlier detection in text I

QUESTIONS AND TASKS:

• Recurrent NN for NLP. Describe one task.

• Outliers in text data.

• Methods for outlier detection in text

MATERIALS FOR THE LABS IN WEEK 06 (all necessary details in the notebook):

LSTM

RNN/LSTM case studies

Michal Hala

Tomáš Houfek

Andrej Betík

Dominik Tuchyňa

Radoslav Sabol

ILP

Poster session. Preliminary program

Neural Code Search: How Facebook Uses Neural Networks to Help Developers Search for Code Snippets

Andrej Betík

XTREME: A Massively Multilingual Multi-task Benchmark for Evaluating Cross-lingual Generalization

Michal Hala

Hierarchical Attention Network

Tomáš Houfek

Deep learning for question answering in Czech

Radoslav Sabol

Transformers

Dominik Tuchyňa

QUESTIONS AND TASKS:

• Why we need LSTM. Describe a task where LSTM areuseful

• ILP. why we need it. Describe a task where ILP is useful

Materials for the labs in week 08:

Preliminary program

Neural Code Search: How Facebook Uses Neural Networks to Help Developers Search for Code Snippets

Andrej Betík

XTREME: A Massively Multilingual Multi-task Benchmark for Evaluating Cross-lingual Generalization

Michal Hala

Hierarchical Attention Network

Tomáš Houfek

Deep learning for question answering in Czech

Radoslav Sabol

Transformers

Dominik Tuchyňa

Learning language in logic

Keyness

Text summarization

QUESTIONS AND TASKS:

• Key words, keyness. How to compute.

• Relational learning  (ILP) for key words (key phrases) detection

• Text sumarization.  Two main approaches.

• Extractive sumarization. ROUGE-n.

Materials for the labs in week 08:

Lecture materials:

Labs (06) materials:

QUESTIONS AND TASKS:

• Sentiment analysis - the definition of the field, justification of its practical relevance and main challenges

• Detailed overview of a selected lexicon-based approach to sentiment analysis

• Detailed overview of a selected classical machine learning approach to sentiment analysis

• Detailed overview of a selected deep learning approach to sentiment analysis

• Comparison of lexicon-based, classical machine learning and deep learning approaches to sentiment analysis

Lecture materials:

QUESTIONS AND TASKS:

• Basic principles of knowledge representation

• Ontologies vs. knowledge graphs - pros and cons of each approach to knowledge representation

• The stack of typical tasks in ontology learning

• Main challenges and open problems of ontology learning

• Techniques used for term extraction, synonym discovery and concept formation

• Techniques used for taxonomy extraction

• Techniques used for relation, rule and axiom extraction

• Overview of a selected deep learning approach to knowledge extraction