Using Textual Semantic Similarity to Improve Clustering Quality of Web Video Search Results

Clustering Web video search results is to help users locating videos of interest in more effective manner. To cluster returned videos, existing works proposed to use textual and visual similarity of videos. However, one of their limitations is that semantic similarity of textual metadata was not considered. Meanwhile, metadata of videos are usually annotated by users with words of high semantic level. This paper introduces a thesaurus based approach to estimate textual semantic similarity of metadata for clustering Web video search results. Experiments were conducted on a set of real-world videos crawled from the Internet. The experimental results demonstrated that using semantic similarity of textual metadata in the combination with visual similarity significantly improves clustering quality.