Website Layout Recommender System

Abstract

Manually creating attractive business landing page in online advertising scenario is both knowledge and labor intensive. In this thesis we present individual building blocks of an Automated Landing Page Generator, namely Color-Scheme Generation, Text Highlighting, Background & Foreground Image Classification and Center-of-Mass balancing of a landing page, given the business description, image, logo provided by the business owner. The Color-Scheme is chosen based on the saturation and value of colors in the HSV-format. For Text-Highlighting we tokenize and tag the input business description and highlight all adjectives, adverbs, numbers and nouns to which they adhere. Background & Foreground Image Classification is based on the presence of unicolor borders or transparent pixels. The unicolor border detection uses thresholding technique to allow indistinguishable color variations. It can be adapted either with RGB color values or with the lightness values in HSL format. Finally, layouts are balanced based on their Center of Mass using a constraint satisfaction solver. To evaluate the proposed RGB and HSL Classifier, we run the classifier on an artificial dataset composed of 148 images. The RGB Classifier achieves better performance for thresholding in terms of robustness against larger thresholds. We also design a user study to evaluate the proposed system. We let users compare in terms of attractiveness, and rate both the original landing page and the generated ones under different parameter configurations. Result shows that automatically generated landing pages attend similar ratings as the original ones.