Open-domain question classification and completion in conversational information search

TL;DR

This paper presents an open-domain conversational search system that improves question understanding and answer retrieval by question completion and classification, achieving 20% more relevant results than baseline methods.

Contribution

It introduces a novel framework for conversational information search that combines question completion and classification, implemented on the TREC CAsT 2019 dataset.

Findings

Achieved 20% more relevant results than baseline methods.

Developed a framework for question completion and classification in conversational search.

Demonstrated high performance with a simple method.

Abstract

Searching for new information requires talking to the system. In this research, an Open-domain Conversational information search system has been developed. This system has been implemented using the TREC CAsT 2019 track, which is one of the first attempts to build a framework in this area. According to the user's previous questions, the system firstly completes the question (using the first and the previous question in each turn) and then classifies it (based on the question words). This system extracts the related answers according to the rules of each question. In this research, a simple yet effective method with high performance has been used, which on average, extracts 20% more relevant results than the baseline.