• Abstract
  • This paper is a comprehensive evaluation of the current state of the art dialogue system techniques and datasets
• Introduction
  • Dialogue systems perform chit-chat with human or serve as an assistant via conversations
  • There are two main types of dialogue system applications
    • Task oriented dialogue systems (TOD)
      • Solve specific problems in a certain domain such as movie ticket booking, restaurant table reserving, etc.
    • Open domain dialogue systems (OOD)
      • Aims to chat with users without task and domain restrictions
      • They are usually data-drive
  • Both dialogue systems can be seen as a mapping $\phi$ from user message $U = \{u^{(1)}, u^{(2)}, \ldots,u^{(i)}\}$ to agent response $R = \{r^{(1)}, r^{(2)}, \ldots,r^{(i)}\}$ where $R = \phi(U)$
  • In many ope-domain and task-oriented dialogue systems, this mapping also considers a source of external knowledge/database $K$ as input
    • such that $R = \phi(U,K)$
• Neural Models in Dialogue Systems
  • Summary
    • Introduce neural models thata re popular in state of the art dialogue systems and related subtasks
  • Convolutional Neural Networks
    • Considered one of the most powerful models
    • ‘Deep’
      • refers to the fact that they are multilayer which extracts features by stacking feed-forward layers
      • Feed forward layers can be defined as
        • $y = \sigma(Wx + b)$
        • where $\sigma$ is an activation function which makes the otherwise linear operation, non-linear
        • $W$ and $b$ are trainable parameters
• Task Oriented Dialogue Systems
  • Overview
    • There are two main types of systems in TOD systems
      • Modular
        • Natural Language Understanding (NLU)
        • Dialogue State Tracking (DST)
        • Dialogue Policy Learning
        • Natural Language Generation
      • End to End
  • NLU
    • The NLU module impacts the whole system significantly in terms of the response quality
    • Converts natural language message produced by the user into semantic slots and performs classification
    • manages three tasks
      • Domain classification and intent detection
        • Both of these are classification problems
        • They use classifiers to predict a mapping from the input language sequence to a predefined label set
          • For example, the predicted domain can be “movie” and the intent is “find_movie”
      • Slot filling