Web Video Clustering based on Cluster Ensemble by using Text Information

Abstract

In social media, web video categorization is the main challenge for organizing and retrieving the number of videos on the web. From this perspective, a novel framework for social web video clustering is proposed by using textual features, e.g., title, tag and description, and external support source based on semantic relation. Firstly, we modify the traditional similarity by adjusted weights of feature vectors in Vector Space Model (VSM). Secondly, by considering the semantic information between terms, the Word2vec is used to capture the continuous vectors of terms in a document. Thirdly, a new distance function for measuring the similarity between two documents is defined based on link-based Semantic Relation of the relevant terms in each category document. Fourthly, the combining function is approached to combine the similarities before applying them to clustering models. Finally, we employ the Clustering Ensemble to integrate the labels of each clustering. Experimental evaluations on real-world web video datasets demonstrate the advantages of the proposed approach.