Why, vitrivr? Understanding Results in Multimedia Retrieval (Master Project, Finished)


Cristina Illi


As the amount of digitally available information is consistently growing, the demand of effective retrieval systems becomes of greater importance in order to make efficient use of the data. To reach a maximum synergy between user and system, a higher transparency of the inner workings and results is vital. This project focused on extending the functionality of vitrivr, an open-source content-based multimedia retrieval stack, in order to create more transparency and therefore bringing the user and the system closer together. A prioritisation feature for the search tags, statistical as well as relational insights on the result set and feature information about individual elements were added to vitrivr. An evaluation was carried out to examine the impact of the added changes on user performance and understanding. The participants using the new version of vitrivr appeared to be more successful. Indicators that the additions made it harder for the participants to use the systems were not found. The added functionality to vitrivr presents a good step towards higher query quality and system transparency through explainability.



Start / End Dates

2020/09/21 - 2021/01/31


Research Topics