korymath / dl-for-chatbot

Deep Learning / NLP tutorial for Chatbot Developers

Deep leaning for Chatbot Developers

• Course Materials of Deep leaning for Chatbot Developers (Sep. 2017)
• Author: Jaemin Cho
• Pull Requests welcome :)

Contents

Day 01 Introduction to Chatbot (slideshare)

• Introduction to NLP/Chatbot
• Overview of Korean/English NLP Toolkits/Datasets
• Tutorial (code)
  • Introduction to spaCy / gensim / konlpy / other Korean toolkits
  • Sentiment classification via TF-IDF (scikit-learn)
  • Chatbot Pipelining / Serving via Kakaotalk (flask) / Slack (slacker)

Day 02 Text Classification with CNN/RNN (slideshare)

• CNN for text classification
  • Word CNN / Dynamic CNN / Char CNN / Very Deep CNN
• RNN for text classification
  • Bidirectional RNN / Recursive NN / Tree LSTM / Dual Encoder LSTM
• Advanced CNN/RNN architectures
  • QRNN / SRU / ByteNet / SliceNet / LSTM-CNNs-CRF
• Tutorial (code)
  • Word-CNN for sentiment analysis
  • PyTorch Style Guide
  • TorchText Tutorial

Day 03 Conversation Modeling with Seq2Seq / Attention (slideshare)

• Seq2Seq models for conversation modeling
  • Seq2Seq / Neural Conversation model / Diversity-prompting objective: MMI
• Advanced Seq2Seq architectures
  • Show and Tell / HRED / VHRED / Personal based Neural Conversation model / Contextualized Word Vectors (CoVe)
• Attention mechanism
  • Bahdanau / Luong
  • Global / Local
• Advanced Attention architectures
  • Show, Attend and Tell / Pointer Networks / CopyNet / BiDAF / Transformer
• Tutorial (code)
  • Seq2Seq with Attention for Machine Translation

Day 04 QA with External Memory (slideshare)

• QA with External Memory
  • Memory Networks / End-to-End Memory Networks / Key-value Memory Networks / Neural Turing Machines
• Advanced Memory architectures
  • DNC / Life-long memory Modules / Context-Sequence Memory Networks
• Advanced Dialogue Architectures
  • MILABOT / Dialog based language learning / End-to-End Goal Oriented Dialog / Deep RL / Adversarial
• Tutorial (code)
  • End-to-End Memory Networks for Question Answering (bAbI)

Dependencies

Python 3

• Codes are written in Anacodna Python 3.6.
• Package management via Conda or virtualenv is recommended.

ML / NLP

• PyTorch
• TorchText
• spaCy
• sckit-learn
• gensim
• konlpy (requires Jpype3)

Interactive / DataFrame / Plot

• jupyter
• pandas
• matplotlib

Kakaotalk / Slack Bot

• flask
• websocket-client
• beautifulsoup4
• slacker