Checkout different approaches

[JohnMrziglod]

• How does a chat bot work?
• Read the links about different implementations of chatbots
• Make a pro/con list

[JohnMrziglod]

ELIZA - http://www.manifestation.com/neurotoys/eliza.php3

• Tries to emulate a conversation by rephrasing the user's input.

[JohnMrziglod]

Deep Learning with seq2seq model

• One uses two RNNs to encode and decode phrases from a large question-answer-database
• Model learns typical answers for questions but can also generate new answers for questions that it have never seen before.
• A blog post about this approach using tensorflow: https://adeshpande3.github.io/How-I-Used-Deep-Learning-to-Train-a-Chatbot-to-Talk-Like-Me

Pros:

• End-to-End approach: No natural language processing necessary
• Generative approach: Instead of having a large predefined question-answer tree and use heuristic methods, it is smart and generates answers by its own (http://www.wildml.com/2016/04/deep-learning-for-chatbots-part-1-introduction/)
• State-of-the-art according to various blogs and articles

Cons:

• Needs training times of 1-2 days on laptops
• Should be trained on a specific domain to deliver a good performance
• Background/Theory might be more challenging than of other approaches