Web video topic discovery and tracking via bipartite graph reinforcement model

Abstract

Automatic topic discovery and tracking on web-shared videos can greatly benefit both web service providers and end users. Most of current solutions of topic detection and tracking were done on news and cannot be directly applied on web videos, because the semantic information of web videos is much less than that of news videos. In this paper, we propose a bipartite graph model to address this issue. The bipartite graph represents the correlation between web videos and their keywords, and automatic topic discovery is achieved through two steps -coarse topic filtering and fine topic re-ranking. First, a weight-updating co-clustering algorithm is employed to filter out topic candidates at a coarse level. Then the videos on each topic are re-ranked by analyzing the link structures of the corresponding bipartite graph. After the topics are discovered, the interesting ones can also be tracked over a period of time using the same bipartite graph model. The key is to propagate the relevant scores and keywords from the videos of interests to other relevant ones through the bipartite graph links. Experimental results on real web videos from YouKu, a YouTube counterpart in China, demonstrate the effectiveness of the proposed methods. We report very promising results.

Learning similarity metrics for event identification in social media

“…Several efforts have focused on extracting high-quality information from social media [1,4,24,27,31]. Recent studies [21,22] showed that social media document tags are accurate content descriptors, and could be used to train a social tagging prediction system.…”

VisualTextualRank: An Extension of VisualRank to Large-Scale Video Shot Extraction Exploiting Tag Co-occurrence

“…Dong et al [18] proposed a method to evaluate tag relevance score by combining the probabilistic relevance score estimation and random walk-based refinement. Especially, Liu et al [9] presented a Web video topic discovery and tracking method via a bipartite graph which represents the correlation between videos and their tags. Actually, their idea is the motivation of this work.…”

“…VTR is an extension of VisualRank [6] with ideas motivated by [9]. In [9], tags and videos are also co-ranked using their correlation to refine their relevance with a specific topic.…”

“…In [9], tags and videos are also co-ranked using their correlation to refine their relevance with a specific topic. However, unlike our work, in [9], relevance of the whole video, not every scene in it, is evaluated and visual features of videos are totally ignored. On the other hand, VisualRank exploits only a visual linkage between images and does not take textual information into account.…”

“…In Sect. 3 the proposed ranking method, which extends [6] and [9], is presented in detail. We then describe how we apply our method to the system of video shot extraction [7] in Sect.…”