Skip to content

Latest commit

 

History

History
147 lines (116 loc) · 5.17 KB

ReadMe.md

File metadata and controls

147 lines (116 loc) · 5.17 KB
Raw

NLP Chatbot project

Keywords: Python, TensorFlow, Deep Learning, Natural Language Processing, Chatbot, Movie Dialogues

Table of contents

  1. Installation
  2. Introduction
    2.1 Goal
    2.2 Results
  3. Project structure
  4. Dataset
  5. Project usage
    5.1 Reformat the raw data .txt files
    5.2 Train the NLP Seq2Seq model
    5.3 Visualize predictions with trained model
    5.4 Chat with the Chatbot AI
  6. Todo
  7. Resources

1. Installation

This project was designed for:

  • Python 3.6
  • TensorFlow 1.12.0

Please install requirements & project:

$ cd /path/to/project/
$ git clone https://github.com/filippogiruzzi/nlp_chatbot.git
$ cd nlp_chatbot/
$ pip3 install -r requirements.txt
$ pip3 install -e . --user --upgrade

2. Introduction

2.1 Goal

The purpose of this project is to design and implement a realistic Chatbot based on Natural Language Processing (NLP).

2.2 Results

alt text

3. Project structure

The project nlp_chatbot/ has the following structure:

  • nlp/data_processing/: data processing, recording & visualization
  • nlp/training/: data input pipeline, model & training / evaluation / prediction operations
  • nlp/inference/: exporting trained model & inference

4. Dataset

Please download the Cornell Movie-Dialogs Corpus dataset , and extract all files to /path/to/cornell_movie_data/. The challenge description can be found on Kaggle .

The dataset consists of 220 579 conversational exchanges between 10 292 pairs of movie characters and involves 9 035 characters from 617 movies, and is thus well suited for realistic chatbot applications.

5. Project usage

$ cd /path/to/project/nlp_chatbot/nlp/

5.1 Reformat the raw data .txt files

$ python3 data_processing/data_formatter.py --data-dir /path/to/cornell_movie_data/

5.2 Train the NLP Seq2Seq model

$ python3 training/train.py --data-dir /path/to/cornell_movie_data/

5.3 Visualize predictions with trained model

$ python3 training/train.py --data-dir /path/to/cornell_movie_data/tfrecords/
                            --mode predict
                            --model-dir /path/to/trained/model/dir/
                            --ckpt /path/to/trained/model/dir/

5.4 Chat with the Chatbot AI

$ python3 inference/export_model.py --model-dir /path/to/trained/model/dir/
                                    --ckpt /path/to/trained/model/dir/
$ python3 inference/inference.py --data_dir /path/to/cornell_movie_data/
                                 --exported_model /path/to/exported/model/

The trained model will be recorded in /path/to/cornell_movie_data/tfrecords/models/seq2seq/. The exported model will be recorded inside this directory.

6. Todo

  • Full training on Colab
  • Add Google Colab demo
  • Add attention
  • Debug training accuracy
  • Add evaluation accuracy
  • Inference model & script
  • Chatbot interface
  • Clean OOP inference
  • Add architecture on ReadMe
  • Add Beam search decoding & random sampling decoding
  • Add Softmax temperature
  • Add complex models
  • Add & compare with statistical baseline
  • Add perplexity
  • Visualize attention
  • Add char level model
  • Add BPE (Byte Pair Encoding)
  • Train on maximizing MMI (Maximum Mutual Information)

7. Resources

This project was widely inspired by:

  • Pytorch chatbot tutorial, Pytorch website
  • Pytorch NLP tutorial, Pytorch website
  • TensorFlow NLP tutorial, TensorFlow website
  • Keras NLP tutorial, TDS
  • Kaggle challenge, Kaggle
  • Sequence to Sequence Learning with Neural Networks, I. Sutskever, O. Vinyals, Q. V. Le, 2014, Arxiv
  • A Neural Conversational Model, O. Vinyals, Q. Le, 2015, Arxiv
  • Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation, K. Cho, B. van Merrienboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, Y. Bengio, 2014, Arxiv
  • Effective Approaches to Attention-based Neural Machine Translation, M-T. Luong, H. Pham, C. D. Manning, 2015, Arxiv
  • Neural Machine Translation by Jointly Learning to Align and Translate, D. Bahdanau, K. Cho, Y. Bengio, 2014, Arxiv