README.md

NLP Modules (1 month)

Outline

S/N	Domain	Estimated Duration
1	Lectures	1 week
2	Tutorials	1 week
3	Assessment	2 weeks

Lectures

Natural Language Processing

• Stanford Natural Language Processing Slides
• Stanford Natural Language Processing Lectures (YouTube)
  • Introduction (1.1)
  • Basic Text Processing (2.1 to 2.5)
  • Minimum Edit Distance (3.1 to 3.3)
  • Language Modeling (4.1 to 4.6)
  • Text Classification (6.1 to 6.9)
  • Sentiment Analysis (7.1 to 7.5)
  • Information Extraction and Named Entity Recognition (9.1 to 9.3)
  • Relation Extraction (10.1)
  • Part-Of-Speech Tagging (12.1 and 12.2)
  • Information Retrieval (18.1 to 18.3)
  • Ranked Information Retrieval (19.1 to 19.5)
  • Semantics (20.1 to 20.5)
  • Question Answering (21.1 to 21.3)
  • Summarization (22.1)
• You can find the backup to the slides and videos here

Word Embeddings

• deeplearning.ai Sequence Models (week 2)

Tutorials

Natural Language Processing

• The following uses NLTK, but other toolkits that cover these topics are acceptable
• Stemmers and Lemmatisation (section 3.6)
• Tokenising and Segmentation (sections 3.7 and 3.8)
• POS Tagging (sections 1 and 2)

Word Embeddings

• Word Embeddings in Python with Spacy and Gensim

Practical Application

• https://scikit-learn.org/stable/tutorial/text_analytics/working_with_text_data.html

Available Toolkits

• NLTK
• scikit-learn
• fastText
• Gensim
• spaCy

Assessment

This section is a hands-on assessment that requires practitioners to attempt the Kaggle Avito Duplicate Ads Detection prediction competition. You are expected to write your own code from scratch using concepts learnt from Kaggle Titanic and NLP methods (e.g. word2vec). Consequently, you'll present your work to your mentor/supervisor.

Kaggle Avito Duplicate Ads Detection

• https://www.kaggle.com/c/avito-duplicate-ads-detection

Recommended features and models

• Avito Duplicate Ads Detection, Winners' Interview: 2nd Place, Team TheQuants
• You can find the backup to the article here

Vaex (alternative to Pandas if it's too slow)

• vaexio/vaex: Out-of-Core DataFrames for Python