Implementation of recommender systems for predicting interesting new items

V sklopu diplomskega dela so na podlagi dveh podatkovnih zbirk trgovin bili razviti trije različni priporočilni sistemi. Najprej sta podrobneje opisani obe podatkovni zbirki za spletno trgovino Instacart in Mercator. V nadaljevanju so predstavljena orodja in uporabljene metode, potrebne za postavitev sistemov. Sledi potek same implementacije priporočilnih modelov, kjer se podrobno obravnava celoten postopek izdelave. Sledi evalvacija oz. predstavitev rezultatov, kjer je ocenjena uspešnost implementiranih modelov. Najbolj uspešna modela sta predstavljala pristopa, ki uporabnikom nove izdelke priporočata na podlagi podobnosti izdelkov. V nadaljevanju je predstavljena tudi primerjava priporočilnih sistemov tega diplomskega dela, s prirejenim priporočilnim sistemom sorodnega dela. Z ozirom na problem priporočanja novih izdelkov uporabnikom, so se bolje obnesli modeli, implementirani v obsegu tega dela. Delo zaključujejo sklepne ugotovitve, ki izpostavljajo morebitne izboljšave implementiranih modelov.Within the scope of this thesis, three different recommender systems were created, each based on two stores\u27s datasets. Firstly, store datasets are described in detail. Then, the tools and techniques used to implement the systems are presented. After that, the thesis describes the process of implementing the systems in detail. Then, evaluation of implemented recommender systems follows, describing their accuracy. The most successful two strategies to recommend new products to users turned out to be based on product similarity. Next, recommender systems, developed in this thesis, were compared to adapted system, created in closely related thesis. Given the nature of the problem, the recommender systems, developed in the scope of this thesis, performed better. Lastly, the thesis is concluded with possible upgrades, which have a potential to make the systems more accurate