An Evaluation of Contextual Suggestion

Abstract

This thesis examines techniques that can be used to evaluate systems that solve the complex task of suggesting points of interest to users. A traveller visiting an unfamiliar, foreign city might be looking for a place to have fun in the last few hours before returning home. Our traveller might browse various search engines and travel websites to find something that he is interested in doing, however this process is time consuming and the visitor may want to find some suggestion quickly.

We will consider the type of system that is able to handle this complex request in such a way that the user is satisfied. Because the type of suggestion one person wants will differ from the type of suggestion another person wants we will consider systems that incorporate some level of personalization. In this work we will develop user profiles that are based on real users and set up experiments that many research groups can participate in, competing to develop the best techniques for implementing this kind of system. These systems will make suggestion of attractions to visit in various different US cities to many users.

This thesis is divided into two stages. During the first stage we will look at what information will go into our user profiles and what information we need to know about the users in order to decide whether they would visit an attraction. The second stage will be deciding how to evaluate the suggestions that various systems make in order to determine which system is able to make the best suggestions.