Image search: an investigation of factors affecting search behaviour of users

A Hamid, R 2017, Image search: an investigation of factors affecting search behaviour of users, Doctor of Philosophy (PhD), Science, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
AHamid.pdf Thesis Click to show the corresponding preview/stream application/pdf; Bytes
Title Image search: an investigation of factors affecting search behaviour of users
Author(s) A Hamid, R
Year 2017
Abstract Searching for images can be challenging. How users search for images is governed by their information need. Nevertheless, in fulfilling their information need, users are often affected by subjective factors. These factors include topic familiarity, task difficulty, relevance criteria and satisfaction. This thesis focuses on three research questions exploring how image information needs together with these factors affect online web users' searching behaviour.
The questions are:
1. How does image information need affect the criteria users apply when selecting relevant images?
2. How do different factors in image retrieval affect users' image searching behaviour?
3. Can we identify image information needs solely from user queries?

In addressing these challenges, we conducted both user studies and proxy log analysis to complement each other. User studies are conducted in a laboratory setting and the needs are artificial, while proxy log captures users' actual needs and behaviour in the wild. The main user study involved 48 students of various disciplines from RMIT University. In the study, we represent image information needs as types of tasks. Data were collected from questionnaires and screen capture recordings. The questionnaire was used to collect data on criteria users find important when judging image relevance and perception on the effects of subjective factors to their searching. Screen capture recordings of their search activities were observed and time stamped to identify and measure search and retrieval behaviour. These measures were used to evaluate the effects of subjective factors on users' image search behaviour.

The results showed in judging image relevance, users may apply similar criteria, however, the importance of these criteria depend on the types of image. Similarly, ratings of users' perception on aspects of performing image search show they were task dependent and that effect of different aspects were related. Users were more affected by familiarity and satisfaction when performing difficult image search tasks. Results of correlation suggest that users' perception of aspects of performing image search did not always correspond with their actual search behaviour. However, for some subjective aspects of user search behaviour, we have identified particular objective measures that correlate well with that aspect.

The examination of users' queries in proxy logs, shows that users search for unambiguous images more frequently compared to conceptual images. Their sessions are short with two to three terms per query. When analysing queries from logs, we are actually making a guess of what users were searching for. However, by examining the way users modify/reformulate their queries may give an indication of their information need. Results show, that users frequently submit new queries or replace terms from their previous query rather than revising the query into more depth or breadth. Similar findings were found when compared with the user study data, whereby users in both settings exhibit similarity in the number of queries, terms and reformulation type.

This thesis concludes that given similar image information needs, ordinary users make relevance judgements similar to specialised users (such as journalists, art historians and medical doctors) despite giving attention to different criteria of relevance. Moreover, only certain measures of search behaviour used in text retrieval are applicable to image retrieval due to the difference in judging the relevance of textual information and image. In addition, visual information needs can be better inferred when analysing series of queries and their reformulation within a search session.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Science
Subjects Information Retrieval and Web Search
Human Information Behaviour
Keyword(s) Web search engine
User study
Relevance criteria
Subjective factors
Search behaviour
Version Filter Type
Access Statistics: 137 Abstract Views, 204 File Downloads  -  Detailed Statistics
Created: Wed, 13 Sep 2017, 09:42:37 EST by Denise Paciocco
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us