Master's Thesis

advertisement
Dynamic Layout Optimization for Newspaper
Web Sites using a Controlled Annealed
Genetic Algorithm
Gjermund Brabrand
H06MMT
Index
•
•
•
•
•
•
•
•
Introduction
Thesis
Research questions
Method
Prototype
Results
Conclusions
Further work
Introduction
• What is layout optimization?
• Finding the best layout for a given purpose
• What is the problem with most newspaper
web site presentations today?
• Static layout
• Oversized
• Space efficiency
• How can newspaper web site presentations
benefit dynamic layout?
Thesis
• A layout generator for newspapers
• Problems
•
•
•
•
Control - interaction
Individual apperance
Supervision
Workflow
Thesis
• A newspaper layout
• consists of rectangles laid out on a surface in a way that
produce no gaps, and looks good.
• The pagination problem
• computerized process by which layout components is
laid out
• Annealed genetic algorithm
• evolutionary algorithm used for search and
optimization problems
Research questions
• RQ1: How can automated layout procedures benefit
a newspaper web site advantageously?
• RQ2: How can article control be implemented in the
algorithm fitness function without loss of
effectiveness and performance?
• RQ3: How well does the fitness function and human
eye correlate in picking out visually approved
layouts?
Research questions
• RQ4: What positive and negative factors will
automated layout in a newspaper web site have on
the user workflow compared to regular news
posting?
Method
• RQ1: A prototype is developed using standard
CMS design with a layout generator
implemented. A group of personel with
relevant experience will compare regular news
publishing layout with the protoype.
Method
• RQ2: The prototype is used to test out
different solutions for article control. Test of
performance and runtime will determine
which solution to use. RQ3 is used to answer
this questions visual performance issue.
Method
• RQ3: An experiment is carried out to check for
correlation between human eye and the
fitness function.
Method
• RQ4: Based on a survey answered by the
prototype test participants we try to uncover
significant changes in prototype workflow
compared with regular publishing systems.
Prototype
• Principle of the prototype
Prototype
Annealed genetic algorithm
• Initial solution (chromosome)
• chrom = [ 2 5 1 9 7 8 3 6 4 ]
• Mutation
• ”A bad solution is often close to a good solution”
• Prevent local optima
• chrom = [ 2 5 3 9 7 8 1 6 4 ]
• Control operators (discussed later)
• Calculate fitness
• Check solution
• If new.fitness < current.fitness  hold
• If new.fitness within acceptance domain  hold
Prototype
• Alt 1
1.
2.
3.
4.
Swap operator
• Alt 2
Initial solution
Chrom = [ 2 5 1 9 7 8 3 6 4]
Mutation
Calculate fitness
Check for size match in better
positions
Typical result:
Chrom = [ 2 3 4 9 7 8 5 6 1]
1.
2.
3.
3.
Initial solution
Chrom = [ 2 5 1 9 7 8 3 6 4 ]
Mutation
Put priority articles first
Chrom = [ 7 3 4 9 2 8 5 6 1]
Calculate fitness
Prototype
Headliner operator
1. Initial solution
chrom = [ 2 5 1 9 7 8 3 6 4 ]
2. Mutation
2. Fixed solution
chrom = [ 3 5 1 9 7 8 2 6 4 ]
3. Calculate fitness
Results
• RQ1: Prototype
•
•
•
•
Dynamic without being accidental
Autogenerated category sites
Choose layout profile
Article control
• Experiment
• Group 1: test of prototype - survey
• Group 2: fitness functino vs. human eye
• Performance
Results
• RQ2: Algorithm performance is maintained
• RQ3: Correlation between fitness function and
human eye
10 participants
vs. 8 random
fitness solutions
Results
• Survey
• 6 participants
• Experience with web publishing systems (CMS)
• Work at large newspaper web sites
• RQ4: Outcome
• Not enough positioning control of individual articles
• ”think design”
• Easy to use
Conclusion
• Not enough control for professional use
• Frequent change of layout
• Positioning control
• Usefull in other areas
• Personalized presentations
• Webshop product presentations
• Smaller newspapers/online publications
• Choice of method
Further work
•
•
•
•
Test in other areas
Linked articles
Expand function gallery
Advertisement support
Download