the evaluation of the appeal of website designs to a target group, with the aid of artificial neural...

7
The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software Milan Tair and Marko Petrović Singidunum University, Serbia (Milan Tair and Marko Petrović, 2015) The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software

Upload: byron-kennedy

Post on 19-Dec-2015

213 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software Milan Tair and Marko Petrović Singidunum

The evaluation of the appeal of website designsto a target group, with the aid ofartificial neural network software

Milan Tair and Marko PetrovićSingidunum University, Serbia

(Milan Tair and Marko Petrović, 2015) The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software

Page 2: The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software Milan Tair and Marko Petrović Singidunum

AboutWe present a method of using an ANN (artificial neural network) to evaluate the appeal of website design and layout to members of target groups.

(Milan Tair and Marko Petrović, 2015) The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software

✓?✗?

Page 3: The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software Milan Tair and Marko Petrović Singidunum

How did we do it?We have identified 12 parameters with a predefined set of values that can describe website layouts and designs.Different combinations of predefined values for each of the parameters define a range of website designs.These parameters include: Website wrapper width, Elements outlined with borders, Overall website colour brightness, Overall website colour saturation, etc.

(Milan Tair and Marko Petrović, 2015) The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software

Page 4: The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software Milan Tair and Marko Petrović Singidunum

How did we get the training data?We have created a dynamic website preview generator, which generates a website design preview (HTML and CSS) based on values of the 12 identified parameters.Previews generated with random values for the 12 parameters ware displayed to survey participants. Participants have marked these designs as good or bad.Participant have evaluated a minimum of 30 designs.

(Milan Tair and Marko Petrović, 2015) The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software

Page 5: The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software Milan Tair and Marko Petrović Singidunum

How did we train the ANN? 1/2We have collected 6295 random generated website design evaluations made by a total of 76 participants between 19 and 35 years of age.The ANN simulation software used was a JavaScript library called Brain.JS, developed by Heather Arthur.The ANN had 13 inputs, three hidden layers with 40, 30 and 10 neurons, respectively and 1 output neuron.

(Milan Tair and Marko Petrović, 2015) The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software

Page 6: The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software Milan Tair and Marko Petrović Singidunum

How did we train the ANN? 2/2Evaluation data was formatted in a JavaScript array of objects, each containing an input element (an array of input values) and an output element (an array of output values).The ANN was of a feed-forward model, trained with a back propagation algorithm. The training stopped after 100,000 iterations and the training error of 0.26%.The training error of 0.1% was targeted.

(Milan Tair and Marko Petrović, 2015) The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software

Page 7: The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software Milan Tair and Marko Petrović Singidunum

Testing and resultsBy assessing evaluation decisions made by the trained ANN for all 36,864 possible combinations of the 12 parameters, we have identified designs most likely to be appealing to members of the target group, as well as conceptual designs the least likely to be appealing to members of both genders of the target group, whose inputs were used for creating the ANN training data set.

(Milan Tair and Marko Petrović, 2015) The evaluation of the appeal of website designs to a target group, with the aid of artificial neural network software