Studienprojekte

Projekte sind ein essenzieller Bestandteil des Curriculums von Mobile Computing. Die Studierenden bekommen die Möglichkeit, das im Zuge ihres Studiums erworbene theoretische Wissen selbst praktisch umzusetzen. Ein sowohl für StudentInnen als auch für Lehrende immer wieder spannendes Unterrichtskonzept, in dem schon erfolgreiche Startups wie z.B. runtastic und Butleroy ihre Anfänge gefunden haben.

User Interaction Based Recommender System for Travel Offers on Mobile Devices

Zeitraum
Oct 2017 - Feb 2018
FH Studierende
Christian Obermayr
FH BetreuerIn
FH-Prof. Mag. DI Dr. Clemens Holzmann

Ziel

This project is about creating a recommender system for an existing product, which allows to access travel offers, gathered from various blogs and websites. The RS can be accessed via an Android Application. First, the user has to interact with some items (travel offers), then he/she can request recommendations via a dedicated tab in the application.

Umsetzung

The technique implemented in this project, is called “Collaborative-Filtering”, which means that recommendations are made based on user-similarities and user-item-interactions. More concrete, a user has to interact with items. After some interactions, the RS is able to calculate similarities between the mentioned user and other users, solely based on how they interacted with identical items. If some similarity is found, additional items can be predicted, again based on how a similiar user interacted with them. The goal is, to only recommend items, which are supposed to be liked by the receiving user.

One major task during the development, is to correctly interpret the recorded user-item-interactions. In this project, two actions have been recorded for this purpose. First, how long a user has viewed an item and second, how often he/she has interacted with it. Interactions in this context, are simple website navigations or clicks on links. Another very important task was, to find a suitable algorithm, for calculating the correlation. The Pearson correlation coefficient and the Spearman’s rank correlation coefficient have been implemented and compared in this project, where the latter led to slightly better results in some cases.

When everything was implemented, the RS had to be evaluated. Therefore, 5 participants have been asked to interact with some items and then rate the recommended items. With using the Spearman’s rank correlation coefficient, the recommendations have been ordered correctly, 3 out of 5 times, whereas the Pearson correlation coefficient only recommended the items in the correct order 2 out 5 times. Also the average error in the predicted rating was lower for Spearman’s rank correlation coefficient (1 vs. 2,3). At all, the Spearman’s rank correlation coefficient did a better job in predicting the ratings and therefore ordering the items correctly.