Users often look for information on YouTube, based on this information they inquire about a topic. My goal is to help users to get acquainted with sights (such as the Great Wall of China) in more detailed and more accurately by the improvement of YouTube hit lists.
My aim was to create a clustering algorithm and a reordering procedure for serving user needs. The algorithms require training set for improving the result lists. The training set is created by annotating person or persons. The original YouTube videos hit list should be segmented for the training set. The annotation consists of the grouping of video file segments and analysis of their relevance to the particular attractions. The typical algorithms need for parameters of video file segments, I have implemented programs for extraction of these parameters. The parameters were different types of histograms, and the results of face detection, the edge detection and circle detection. The clustering and reordering algorithms are implemented, which use the results of computed parameters at the examining of an element.
I have achieved improvement in the result list, which is characterized by diversity and precision indicators.