pradip1307 / chatbot-1

Victor- A generative ChatBot created using sequence-to-sequence model consists of two recurrent neural networks (RNNs): an encoder that processes the input and a decoder that generates the output.It can be trained on any desired dataset for specific purposes. Instead of ordinary ChatBots which are based on hard-coded responses, it can understand context and respond accordingly.

ChatBot

Victor- A generative ChatBot created using sequence-to-sequence model consists of two recurrent neural networks (RNNs): an encoder that processes the input and a decoder that generates the output.It can be trained on any desired dataset for specific purposes. Instead of ordinary ChatBots which are based on hard-coded responses, it can understand context and respond accordingly.

Description

The fame of Deep Learning is at its epitome today. This here is a simple demonstration of what elementary level Deep Learning can achieve. Our project, Victor is a generative chatbot that is created using sequence-to-sequence model consists of two recurrent neural networks (RNNs).

"Victor" represents one of the several possibilities that are possible by training a sequential neural network. Going into the specifications, it is constructed of three layers of the sequential neural network each containing 128 neurons. The encoder and decoder have a vocabulary size of 20,000 each. The corpus used for training is a collection of dialogues from 617 movies containing about 220,000 conversations (Cornell Movie Dialogue Corpus).

The real "catch" about Victor is that it can be trained on any data we like, it can learn however we want it to. It gives contextual answers instead of hard-coded responses. It draws information from previous conversational exchanges and is thus, closer to how humans interact. The ability to be trained on any data is not only limited to English, it can be trained in any language we want it to any that is the real beauty of it.

The bot performs extremely well on casual conversations that are prevalent in movies given the fact that we had a relatively short time to train it and thus, several parameters had to be compromised (number of neurons in each layer etc.). Talking about future improvements, we can train it on even a bigger dataset and use more layers or more number of neurons in each layer. Also, we could add the feature of training it simultaneously while using it which would further increase its accuracy with experience.

Tools and Libraries used

• TensorFlow
• Tkinter
• PyCharm
• SublimeText

Instructions to run

There are three steps involved in running the chatbot:

To train the bot

1. Open the file "seq2seq.ini".
2. Set 'mode = train'.
3. Run the file execute.py using the code "python execute.py".

To test the bot after training

1. Open the file "seq2seq.ini".
2. Set 'mode = test'.
3. Run the file execute.py using the code "python execute.py".

Finally to launch the chatbot in serve mode run "host.pyw" file.

And, now you can enjoy the fun conversation with the chatbot !!