To refer to or to cite this work, please use the citation to the published version:De Meester, B., Verborgh, R., Pauwels, P., De Neve, W., Mannens, E., and Van Abstract Thanks to ubiquitous Web connectivity and portable multimedia devices, it has never been so easy to produce and distribute new multimedia resources such as videos, photos, and audio. This ever-increasing production leads to an information overload for consumers, which calls for efficient multimedia retrieval techniques. Multimedia resources can be efficiently retrieved using their metadata, but the multimedia analysis methods that can automatically generate this metadata are currently not reliable enough for highly diverse multimedia content. A reliable and automatic method for analyzing general multimedia content is needed. We introduce a domain-agnostic framework that annotates multimedia B. De Meester ( ) 路 R. Verborgh 路 E. Mannens 路 R. Van resources using currently available multimedia analysis methods. By using a three-step reasoning cycle, this framework can assess and improve the quality of multimedia analysis results, by consecutively (1) combining analysis results effectively, (2) predicting which results might need improvement, and (3) invoking compatible analysis methods to retrieve new results. By using semantic descriptions for the Web services that wrap the multimedia analysis methods, compatible services can be automatically selected. By using additional semantic reasoning on these semantic descriptions, the different services can be repurposed across different use cases. We evaluated this problem-agnostic framework in the context of video face detection, and showed that it is capable of providing the best analysis results regardless of the input video. The proposed methodology can serve as a basis to build a generic multimedia annotation platform, which returns reliable results for diverse multimedia analysis problems. This allows for better metadata generation, and improves the efficient retrieval of multimedia resources.