Yarı-otomatik çok-kipli internet içeriği çeviri sistemi

Non-desktop platforms such as mobile phones, tablets, and connected TVs now serve as primary means for browsing the web. Unfortunately, much of the existing web content has been designed and optimized for desktop viewing. Hence, tailoring the content for viewing on multiple platforms with various constraints has become a necessity. However, manual customization for each platform is costly, and not always feasible. In this work, we introduce a semi-automatic multimodal web content retargeting system that works with minimal assistance of an expert. From a single page, our system learns how a web portal should be tailored based on verbal and gestural descriptions of a human operator, and converts previously unseen pages based on what it has learnt. We have three contributions. First, we introduce a modular system architecture with a flexible backend for learning content translation rules from examples using machine learning. Our second contribution is a translation engine capable of applying previously learned translation rules to translate unseen pages. Our third contribution is a large database consisting of 300 manually coded web pages from 10 popular portals, which will serve as a benchmark database for evaluating similar systems in the future. Our discussion is supported by objective evaluation of the system's ability to learn, and results from a carefully designed usability study demonstrating the effectiveness of the proposed system with real users. The results show that our system can learn conversion rules and it is a usable system.