[Pycon] [new paper] "Bhavani Ravi" - Machine Learning behind chatbots

[info a pycon.it]

Title: Machine Learning behind chatbots
Duration: 45 (includes Q&A)
Q&A Session: 15
Language: en
Type: Talk

Abstract: **Description**

Google Assistant and Siris’ of the world have tickled our curiosity enough to deep dive and understand under the hood technologies that make a chatbot.
Though we don’t have Google level of data to create a generalized chatbot, we can use the existing NLP engines and create chatbots that produce valuable results in a specific domain.

This talk is structured from my quest to build a chatbot engine and what I understood by demystifying an opensource chatbot engine.

**Audience**
The talk is for someone who is curious about chatbot technologies and want to get a deeper understanding how they work. A bit of python and a few ML beginner videos are enough to get you started building bots.

At the end of the talk, you will have an understanding of how chatbot engines work and how to tackle some of the challenges. 

**Outline**
1. Chatbot’s architecture (5 mins)
          In this section, I will introduce the chatbot and technologies involved in building one. We will discuss how building a chatbot is different from building a web application. I will take a sample site and walk them through what structural changes needed to convert the site into a chatbot 

2. ML behind Chatbots (16 mins)
 Introduction to Natural Language Processing(NLP) which includes Natural Language Understanding (NLU) and Natural Language Generation(NLG)
Intent Classification - an essential component of understanding user requests 
Implementation of Intent Classification - tensorflow embedding vs SVM
Entity Extraction - NER
How the entire processing pipeline combines these ML components with other components of a chatbot.

3. Q&A - 4 mins



Tags: [u'machine-learning', u'chatbot', u'nlp']