Paper Info
Title
Multi-turn QA: A RNN Contextual Approach to Intent Classification for Goal-oriented Systems.
Abstract
QA systems offer a human friendly interface to navigate through knowledge, which can range from encyclopedic to domain-specific. Generally, a QA system is designed to provide an answer to a specific question once (so-called single turn) and state-of-the-art systems reach nowadays robust performance in such a scenario. However, most of the interactions with QA systems are based on multiple handshakes of question/answer pairs, where the human being refines the questions further, while the system can collect the necessary information and generate a compelling final answer through multiple turns. In this paper, we investigate and experiment a multi-turn QA system that is suited to work given a particular domain of knowledge and configurable goals. Our approach models the entire dialogue as a sequence of turns, i.e. questions and answers, using a Recurrent Neural Network which is firstly trained to understand natural language, classifying entities and intents using prior knowledge of domain-specific interactions, and provide answers according to the domain used as background knowledge. We have compared our approach with state-of-the-art sequence-based intent classification using a well-known and standardized gold standard observing an increase of 17.16% of F1. Results show the robustness of the approach and the competitive results motivate the adoption in multi-turn QA scenarios.
Year
DOI
Venue
2018
10.1145/3184558.3191539
WWW '18: The Web Conference 2018 Lyon France April, 2018
DocType
ISBN
Citations 
Conference
978-1-4503-5640-4
0
PageRank 
References 
Authors
0.34
12
3
Name
Order
Citations
PageRank
Martino Mensio103.72
giuseppe rizzo2132.29
Morisio, Maurizio31342119.37