Spying out user preferences
The World Wide Web was regarded by scientists working on the project EVALUATE (Theory and practice of algorithms for analysis of people and data on the web) as the largest database available. This collection of data coming from social networks, e-commerce systems and e-government – just to name a few – is so broad and complex that it is difficult for users to find the desired information. Adapting a search engine to cater to specific user needs is still an open research problem that the EVALUATE team sought to address. Users reveal their preferences through the choices they make. These choices may be the click on a particular link in a web search ranking. Previous studies employed users' explicit feedback to adapt the search engine ranking. Users are, however, usually unwilling to give explicit feedback, making the feedback received too limited to be representative. To overcome this problem, the researchers developed active learning algorithms using implicit feedback data to optimise the ranking functions. In contrast to traditional (passive) learning, the new tools may collect a pool of instances and then choose a subset for which data are available. EVALUATE researchers focused on two applications: learning to rank pairwise preferences and clustering with additional information coming in the form of pairwise constraints. The goal was to order linearly elements from the most to the least preferred while disagreeing with as few pairwise preferences or constraints as possible. The new techniques were proven to outperform previous algorithms. The most popular online applications, including search engines and social networks, depend on the ability to provide query results, ads and recommendations. Research work carried out during the EVALUATE project goes well beyond the traditional borders of computer science. The multidisciplinary approach adopted to problems of massive data promises to bring humans into the data analysis loop at all stages.
Keywords
User preferences, online resources, World Wide Web, algorithms, social networks, search engine
