Predicting Links for Mobile GUI Prototyping

No Thumbnail Available

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2022-08-22

Department

Major/Subject

Human-Computer Interaction and Design

Mcode

SCI3020

Degree programme

Master's Programme in ICT Innovation

Language

en

Pages

54+11

Series

Abstract

Creating and maintaining links in prototypes of graphical user interfaces (GUIs) is a manual and often error-prone process which demands considerable resources from designers. Implementing a system that automates or assists this recurring task could allow designers to revise and iterate their prototypes more frequently. In an effort to progress towards such a system, this work develops seven link prediction models based on common GUI embedding techniques---two utilizing online learning and five utilizing heuristics---to generate suggestions for links in mobile GUI prototypes. The link prediction models are trained and tuned using a large-scale dataset of mobile application links (56,790 screen pairs from 5,362 Android applications). They are then tested on a small-scale custom dataset of mobile GUI designs (16 screens across 4 applications) and evaluated using a quantitative online end-user study (n = 36) rating a sample of links from those designs. Using live application data, six out of seven models outperform a random estimator, the online learning-based classifiers fail, however, to outperform the heuristic models. Using the custom small-scale set of GUI design data, all models achieve insufficient performance retrieving links chosen by professional interaction designers. Although the vision of an assistive or automated system for GUI prototyping is enticing, better performing models will be required for its realization.

Description

Supervisor

Oulasvirta, Antti

Thesis advisor

Barz, Michael

Keywords

link prediction, mobile GUIs, GUI prototyping, online machine learning

Other note

Citation