Recent Applied Techniques for Open Dialog Generation Systems

Abstract

Dialog Generation Systems (DGS) have emerged as a critical aspect of Natural Language Processing in recent years (NLP). It enables a diverse set of relevant applications to interact with humans in a natural and intelligent way. This study provides a systematic review of open DGS techniques that have recently been used. The major goal of this study is to discuss and analyze the most widely used approaches for implementing DGS's that have been published in recent years. Also, the most popular datasets for open DGS are enumerated, and some commonly used automatic evaluating metrics are presented. As a result, the explored methods are categorized into six main categories, Reinforcement Learning (RL), Hierarchical Recurrent Encoder-Decoder (HRED), Generative Adversarial Networks (GAN), Variational Auto-Encoder (VAE), Sequence to Sequence (Seq2Seq), and Pretraining Model.