Natural Language Understanding in Smartdialog: A Platform for Vietnamese Intelligent Interactions

Abstract

Nowadays in the modern world, interactive smart dialogs with text or voice are gaining traction as the main digital interaction channel between human and machine. However, most of the current platforms do not support or have not fully developed for Vietnamese. In this paper, the authors propose a smart conversational platform through a text channel and/or voice channel in Vietnamese language, including these main steps: (i) Input Conversion and Pre-Processing, (ii) Entity Recognition, (iii) Intent Classification, (iv) Action Prediction and Execution, and (v) Output Generation. This paper focuses on presenting problems related to natural language understanding. To recognize entities in a sentence, the authors studied and optimized the features for Vietnamese with the Conditional Random Field model. With the problem of predicting user intent, this work proposed, experimented, and compared of Random Forest and BiLSTM deep learning model to optimize for the Vietnamese language. A platform was built and deployed for Milo smart speaker application (LUMI smart home) and VADI driver virtual assistant with the accuracy of around 98.7%.