BOON : An Intelligent Chatbot using Deep Neural Network

Description

• Developed chatbot using encoder and decoder based Sequence-to-Sequence (Seq2Seq) model from Google’s Neural Machine Translation (NMT) module and Cornell Movie Subtitle Corpus.
• Seq2Seq architecture built on Recurrent Neural Network and was optimized with bidirectional LSTM cells.
• Enhanced chatbot performance by applying Neural Attention Mechanism and Beam Search.
• Attained testing perplexity of 46.82 and Bleu 10.6.
• Developed backend using Python and front-end using Python and PyQT.

Sample Conversations:

GUI:

Requirement:

• Google's Tensorflow 1.6
• Python 3.4
• tensorboard

Instruction For running chatbot:

• Activate tensorflow
  source ~/tensorflow/bin/activate
• Move to folder "project/code/"
  cd /home/rkc007/projects/dode/
• python chat_gui.py

Training

1. Activate tensorflow

   source ~/tensorflow/bin/activate

2. move to nmt module: change prefix "/home/rkc007/project" to current project location:

   cd /home/rkc007/project/code/nmt

3. remove model folder: change prefix "/home/rkc007/project" to current project location

   rm -r /home/rkc007/project/code/inout/output/nmt_model

4. create nmt training output folder: change prefix "/home/rkc007/project" to current project location

   mkdir /home/rkc007/code/inout/output/nmt_model

5. update location of "vocab_prefix", "train_prefix", "dev_prefix", "test_prefix", "out_dir" by changing the prefix in all these files with where prefix "/home/rkc007/project" should be changed to current project folder. Also, have option to change - "num_train_steps", "steps_per_stats", "num_units", "learning_rate", "beam_width", "num_translations_per_input"

Following is the training instruction :

python -m nmt.nmt \
    --src=vi --tgt=en \
    --vocab_prefix=/home/rkc007/project/Code/inout/input/nmt_data/vocab  \
    --train_prefix=/home/rkc007/project/code/inout/input/nmt_data/train \
    --dev_prefix=/home/rkc007/project/code/inout/input/nmt_data/tst2012  \
    --test_prefix=/home/rck007/project/code/inout/input/nmt_data/tst2013 \
    --out_dir=/home/rkc007/project/code/inout/output/nmt_model \
    --attention=scaled_luong \
    --num_train_steps=12000 \
    --steps_per_stats=100 \
    --num_layers=2 \
    --num_units=512 \
    --learning_rate=0.001 \
    --decay_steps=1 \
    --start_decay_step=1 \
    --beam_width=10 \
    --length_penalty_weight=1.0 \
    --optimizer=adam \
    --encoder_type=bi \
    --num_translations_per_input=30

For viewing training performance:

1. Activate tensorflow
   source ~/tensorflow/bin/activate

2. move to nmt module:
   cd /home/rkc007/project/code/nmt

3. enter following command
   tensorboard --port 22222 --logdir /home/rkc007/project/code/inout/output/nmt_model

Future Work:

1. Working on creating a server using Flask to host this chatbot.
2. Chatbot to work on public domain with different languages

Acknowledgements:

Major Thanks to Dr. Kevin Scannell for teaching NLP course.
May Li
TensorLayer Community
Aditya Kumar for GUI