• Course's URL: https://www.coursera.org/learn/language-processing/home/welcome
• Official github repo of the course: https://github.com/hse-aml/natural-language-processing
• Teach by : Anna Potapenko
• Project: build a chat-bot
• TOC

Setting up

• Locally: using docker: https://github.com/hse-aml/natural-language-processing/blob/master/Docker-tutorial.md

  docker pull akashin/coursera-aml-nlp
  docker run -it -p 8085:8080 --name coursera-aml-nlp -v $PWD:/root/coursera --gpus all akashin/coursera-aml-nlp
  
  # inside container "coursera-aml-nlp", run the notebook
  run_notebook
  # then go: <http://localhost:8085/tree/coursera>
  
  # stop container
  docker stop coursera-aml-nlp
  
  # start container
  docker start -i coursera-aml-nlp
  

Brief overview

• Week 1: text classification tasks

  • predict some tags or categies
  • predict sentiment for a review
  • filter spam e-mails.

  ⇒ Bag of words

• Week 2: predict word sequences + predict tags for word sequences

  • Language modeling: predict the prob of the next word given some previous words.
    • used in text generation. chatbots, speech recognition, machine translation, summarization,...
  • Sequence tagging: have a sequence of words and need to predict tag for each of word in the sequence.
    • part-of-speech tags: predict nouns, verbs,... of the words
    • named entities: find some names or entities and use them as features for other tasks
    • semantic slots: about some slots → eg. the person wanna book a table for some specific time or specific place ← those time and space would be the slots!

• Week 3: How to represent a meaning of a word, a sentence or a text? ⇒ put them in vectors

  • Word embeddings ← similar words have similar vectors

    

  • Sentence embeddings

  • Topic model ← deal with whole document and represents them by some vectors → give the topic of certain document! → eg.

• Week 4: Sequence to sequence task:

  • Machine translation
  • Summarization, simplification.
  • Conversational chat-bot

• Week 5: build a dialogue agents

  • goal-oriented (eg. help in a call-center)
  • conversational (eg. entertainment)
  • Project: build a chatbot that assists with StackOverflow search.

Week 1 - Intro and text classification

Intro to NLP and the course

Intro about the course

• Who is this course for?
  • In-depth theoretical insights
  • Review of state-of-the-art approaches (both in reserach and production)
  • Gaining intuition
  • Practice + project of building a chat-bot.
• This course is not for you if:
  • want only practical tips + don't care what's inside.
  • don't have any prior background in machine learning.

Main approaches in NLP: