| Popis: |
Profiliranje uporabnikov postaja vse bolj pomembna tema pri razvoju spletnih strani, saj omogoča zagotavljanje boljše uporabniške izkušnje z ugotavljanjem uporabnikovih interesov. V tem delu se ukvarjamo z dinamično izbiro metod profiliranja uporabnikov. Cilj je uporabiti metode strojnega učenja in zgraditi učni model, ki bo znal kar najbolje kombinirati metode profiliranja in tako ustvariti kombinirano metodo profiliranja, uspešnejšo od vsake posamezne metode, ki smo jih uporabili pri učenju. Pokazali smo, da je kombiniranje profilirnih algoritmov z uporabo strojnega učenja lahko močno orodje pri izboljšavi uspešnosti profiliranja. Pokazali smo tudi, da je uporaba dinamične izbire metod smiselna v primeru, ko so razlike med posameznimi algoritmi profiliranja večje in so tako tudi možnosti za izboljšave večje. User profiling is becoming an increasingly important subject in the field of web development as it enables improving the user experience through learning the users interests. In this study we examine dynamic selection of web user profiling methods. Our goal is to use machine learning methods to build a learning model that predicts the most successful combined profiling method, which is expected to be significantly better from each individual method. We have shown that combining of profiling methods using machine learning can be a powerful tool when looking for a way of improving the accuracy of web user profiles. We have also shown that dynamic selection is most effective when differences between profiling methods are relatively large and therefore providing room for improvement. |
