Evaluating Algorithms for Temporal Queries in Ad-Hoc Video Retrieval (Bachelor Thesis, Finished)


Viktor Gsteiger


Expressing a temporal relationship between different search queries has become more important in recent years, especially when working with extensive video and audio data collections. Enabling queries with such a relationship is achieved by using temporal queries processed by temporal scoring algorithms. These algorithms aggregate the result sets of multiple search queries according to a temporal relationship and score the results regarding the similarity to the temporal query. In this thesis, seven such algorithms were developed and evaluated regarding response time and searched-item ranking as the primary metrics using a dataset of 109 queries specifically developed to test temporal query algorithms and specified in a newly developed format. The two best-performing algorithms were subsequently implemented in vitrivr, a multimedia retrieval system, with changes to both the front-end and the back-end. The implementation was afterwards successfully used during a competitive evaluation of interactive multimedia retrieval systems. The competitive evaluation has shown that temporal querying in vitrivr has noticeably improved with regards to response time and searched-item ranking due to the new algorithms and the new implementation within vitrivr.



Start / End Dates

2021/03/01 - 2021/06/30


Research Topics