modeling information navigation : implication for information architecture craig s. miller 이주우

Modeling Information Navigation :

Implication for Information Architecture

Craig S. Miller

이주우

INTRODUCTION

MODELING INFORMATION NAVIGATION

SIMULATIONS

COLLECTING DETAILED HUMAN PERFORMANCE RESULTS

DETAILED SIMULATIONS AND COMPARISONS

GENERAL DISCUSSION

APPENDIX

CONTENTS

INTRODUCTION

The identification of factors that affect the usability of the World Wide Web has become increasingly important

How a site’s information architecture impacts a user’s ability

Our goals is

- To show the viability of a computational model

- Advice to Web site designers

-Use these costs identifying effective information architectures

MESA (Method for Evaluating Site Architectures)

MODELING INFORMATION NAVIGATION

Our goal is to simulate common pattern of user interaction with Web site

Guide abstraction and model construction

-The limited capacity principle

-The simplicity principle

-The rationality principle

2.1. Representing a Web Site

2.2. Modeling the Browser and User Actions

Selecting a link (and attending to and identifying a new page)

Pressing the Back button (and attending to and identifying a new page)

Checking a link and evaluating its relevance

2.3. Modeling Navigation Strategies

MESA navigates a Web site by serially executing

- The threshold strategy

- The comparison strategy

- The opportunistically strategy

2.3. Modeling Navigation Strategies

2.4. Detailed Example

2.3. Modeling Navigation Strategies

2.4. Detailed Example

19 link evaluation

8 link selection

5 back action

2.4. Detailed Example

Cognitive limitations contribute to additional navigation cost

-first : only one link can be evaluated at a time

-second : page C needs to be scanned a second time

-third : memory limitation

2.6. Simulation Parameters

Time Constants

Our model estimate navigation times for finding an item

(link evaluation, link selection, and pressing the Back button)

Result from hierarchical menu selection studies. D.P.Miller(1981) reported that humans searching through an 8x8 menu system

Link evaluation / link selection

250 / 500 : 2.9sec

250 / 250 : 2.4sec

250 / 750 : 3.4sec

2.6. Simulation Parameters

Setting Relevance Values for Link Labels

The comparison to menu selection results assumes ideal link. BUT…

- g is produced from a standard normal distribution

- n is noise factor

3. SIMULATIONS

3.1. Modeling Structure and Link Ambiguity

First, explore the effect of label ambiguity on the structure

- Ambiguity link needs more time

- 8x8x8 architecture is faster→ slower

- noise level 0.4

3.2. impact of Time Costs

Although changing the time costs will affect absolute simulation time

-regardless of the time cost

3.3. Impact of Bottom-Level Noise

4. COLLECTING DETAILED HUMAN PERFORMANCE RESULTS

4.1. Method

- Participants : 45, at least 10hr, at least 18year, frequently used the Web

- Materials : 6 high level, 37 low level categories

4.2. Results

Simulation

45 participants Three Structure

Looking for a total of eight items

360 task(15X8X3)

Parameter

500msec : link selecting and pressing the back button

250msec : evaluating each link

Simulation 1 Simulation 2 Simulation 3

ParameterReliability : 0, 0.5, 1Threshold : 0.75 , 0.25

Adding variation to the judges’s average rating

Incrementally increased the

variance

Spearman rank

Correlation0.739 0.851 0.790~0.832

Pearson Correlation

0.692 0.790 0.851~0.869

Simulation

Optimal number of 8 selections per display

12~32 links per page (Larson and Czerwinski(1998))

Two-tired structures produce faster results than Three-tiered in followed three conditions

Three-tired structure may be optimal when the level of label quality is the same across all levels.

Discussion