Dialogue Systems Overview
Dialogue systems are a very appealing technology with an extraordinary future, they are often described as one of the most advanced and promising expressions of interaction between humans and machines. The recent development in deep learning techniques makes the creation of a realistic conversation system, such as a personal assistant or a virtual friend, no longer an illusion. According to the applications, dialogue systems can be categorized into two groups - task-oriented and general conversation systems, known as chatbots.

Task-oriented Dialogue System
Task-Oriented Dialogue Systems help users achieve their specific goals, focusing on understanding users, tracking dialogue states, and generating next actions. The typical structure of a pipeline based task-oriented dialogue system consists of:
  • Natural Language Understanding parses the user utterance into predefined semantic slots
  • Dialogue state tracker manages the input of each turn along with the dialogue history and outputs the current dialogue state.
  • Dialogue Policy defines the rules of transactions to next dialogue state
  • Natural language generation maps the selected action to its surface and generates the response.
Chatbots focus on conversing with humans on open domains. There are two completely different approaches to implementing chatbots:
  1. Retrieval-based model – For a retrieval-based model, we use a corpus of predefined responses and match them for new contexts based on similarity with our responses
  2. Generative model – A generative model chatbot doesn't use any predefined corpus and can generate completely new sentences.

Generative models are able to generate more proper responses that could have never appeared in the corpus, while retrieval-based models enjoy the advantage of informative and fluent responses because they select a proper response for the current conversation from a repository with response selection algorithms. The aditional advantage of using a retrieval-based model is the fact that all unsuitable responses because of grammatical or ethical issues can be easily removed. However, your model will be limited only to the responses defined in the corpus. Generative models, oppositely, can adapt better to surprising demands and questions from customers.

© All Right Reserved. Columb Labs Inc.
e-mail us: info@columblabs.com