The development and assessment of the semantic fields model of visual salience.

Abstract

The present thesis describes the development and assessment of the Semantic Fields Model of visual salience. The Semantic Fields model provides estimates of visual salience in relation to goal-oriented Web site search tasks. The development and assessment of this model is reported over seven studies that are presented in two journal articles and two peer-reviewed conference papers. In Paper 1 (N=50), pupil dilation is validated as a measure of cognitive load for use in later studies. While it has been found previously that a participant’s pupil dilation will be larger during more complex tasks, these experiments have not generally been conducted under the environmental condition of light radiated from a computer monitor. The findings of this experiment indicate that computer monitor radiance in our experimental setting did not interfere with the ability to discriminate successfully between task-related pupil dilation. Paper 2 (N=49) introduces the Semantic Fields model for estimating the visual salience of different areas displayed on a Web page. Latent Semantic Analysis and the Touchstone Applied Science Associates (TASA) corpus were used to calculate Semantic Field values for any (x, y) coordinate point on a Web page based on the structure of that Web page. These Semantic Field values were then used to estimate eye-tracking data that was collected from participants’ goal-oriented search tasks on a total of 1842 Web pages. Semantic Field values were found to predict the participants’ eye-tracking data. In Paper 3 (N=100), four studies are present in which improvements are made to the semantic component of the Semantic Fields model. Estimates of textual similarity generated from six semantic models were compared to human ratings of paragraph similarity on two datasets. Results suggest that when single paragraphs are compared, simple non-reductive models (word overlap and vector space) can provide better similarity estimates than more complex models (Latent Semantic Analysis, Topic Model, Sparse Non-negative Matrix Factorization, and the Constructed Semantics Model). Various methods of corpus creation were explored to facilitate the semantic models’ similarity estimates. Removing numeric and single characters, and also truncating document length improved performance. Automated construction of smaller Wikipedia-based corpora proved to be very effective even improving upon the performance of corpora that had been chosen for the domain. Model performance was further improved by augmenting corpora with dataset stimulus paragraphs. In Paper 4 (N=49), ten models are compared in their ability to predict eye-tracking data that was collected from participants’ goal-oriented search tasks on a total of 1809 Web pages. Forming the basis of six of these models, three semantic models and two corpus types are compared as semantic components for the Semantic Fields model. Latent Semantic Analysis, Sparse Non-Negative Matrix Factorization, vector space, and word overlap were used to generate similarity comparisons of goal and Web page text in the semantic component of the Semantic Fields model. Vector space was consistently the best performing semantic model in this study. Two types of corpora or knowledge-bases were used to inform the semantic models, the well known TASA corpus and other corpora that were constructed from the Wikipedia encyclopedia. In all cases the Wikipedia corpora out performed the TASA corpora. The noncorpus based Semantic Fields model that incorporated word overlap performed more poorly at these tasks. Three display-based models were also included as a point of comparison to evaluate the effectiveness of the Semantic Fields models. In all cases the corpus-based Semantic Fields models outperformed the solely display-based models when predicting the participants’ eye-tracking data. Both final destination pages and pupil data (dilation) indicated that participants’ were actively performing goal-oriented search tasks. Based on this research, it is concluded that the Semantic Fields model provided useful estimates of visual salience during participants’ goal-oriented search of Web sites.