What is it about?
Users of dialogue systems are often limited by the internal knowledge of the system and whenever they ask a question going beyond this internal knowledge base they don't get an answer. Therefore, enabling the access of external knowledge which could provide answers enables user to enter a wider range of questions. To this end, we have integrated a Question-Answering (QA)-module in an interactive restaurant search system and evaluated its impact using a crowd-sourced user evaluation. The results show that systems with QA-module are significantly preferred over the baseline without QA-module.
Featured Image
Photo by Priscilla Du Preez on Unsplash
Why is it important?
Our results show that the dialogue systems with Question-Answering module are significantly preferred over the baseline without such a module. Moreover, we also show that a method to prevent responses unrelated to a user question (‘off-topic responses’) is helpful to reduce the number of off-topic responses in case the question is not covered in the knowledge base, but users showed more preference to the system that can retrieve information irrespective of whether it is relevant or not.
Perspectives
This article combines a previously published system by my co-authors with the additional Question-Answering module which uses separately fine-tuned Large Language Models (GPT-2 ) for both knowledge retrieval and response generation and was a great way to see the level of performance of these models and how they can benefit a modular dialogue system.
Norbert Braunschweiler
Read the Original
This page is a summary of: Enabling Semi-Structured Knowledge Access via a Question-Answering Module in Task-oriented Dialogue Systems, July 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3571884.3597138.
You can read the full text:
Contributors
The following have contributed to this page
