Date of Award

1-1-2020

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

College/School/Department

Department of Information Science

Content Description

1 online resource (xii, 132 pages) : illustrations.

Dissertation/Thesis Chair

Xiaojun Yuan

Committee Members

Hemalata Iyer, Pradeep Atrey, Chirag Shah

Keywords

Automatic speech recognition, Voice computing, Ambient intelligence, Question-answering systems, Query languages (Computer science), Conversation analysis, Dialogue analysis, Human-computer interaction, Information retrieval, Information Storage and Retrieval, information retrieval

Subject Categories

Library and Information Science

Abstract

Conversational search has been obtaining increasing attentions in the field of Information Retrieval (IR). However, compared with the other applications of dialogue systems, the studies on conversational search are far from being sufficient. Lot of work is needed from the input to the output of the conversational search process. On the other hand, with the improvement in the Automatic Speech Recognition (ASR) technology, voice is now an optional input method in addition to text. Compared with the text-based queries, voice-based queries are found to be longer, and use richer language. The voice-based search is also found to be more flexible and complex. When investigating the voice-based search, some lessons could be borrowed from the previous research of text-based search. In this dissertation project, I applied the query reformulation types concluded from text-based search to investigate the partial query modification in voice search. The previous studies indicated that query reformulations are common in voice search for various reasons. Different from the query reformulation in text-based search, where the users can edit the query with keyboard and mouse, the operations used to complete query reformulation in voice search need to be supported by the voice search systems. However, few studies have been performed on voice query reformulation from the IR point of view. This dissertation project aims at filling this gap by investigating 1) how the users would perform partial query modification in voice search and 2) whether the partial query modification in voice search could improve users’ search experience. The dissertation project was composed of three phases, a preliminary survey study, an exploratory experiment, and a usability experiment. The findings of the survey study indicated that the lack of editing function was perceived by the users as one of the limitations of the existing voice search systems. In the exploratory experiment, the participants performed query reformulations on a Wizard-of-Oz system. Three main strategies and several specific operation patterns were concluded from the experimental results. Then, three specific operation patterns were implemented in an automatic system and the usability of the system was tested in the second experiment. The usability experiment revealed that the experimental system led to better user experience in general, compared with a baseline system on which only complete queries were accepted. The findings of the dissertation project can provide some guidance on the design of voice search systems. To avoid complexity, the settings of the experiments introduced some limitations, which are addressed in conclusion.

Download

Share

COinS