Combining Multi-Modal Features for News Story Correlation Analysis
How to combine multi-modal features effectively is a difficult problem in news story correlation analysis, this paper puts forward a new two-stage fusion approach based on visual and textual features fusion to solve this problem. First we use a co-clustering method to get the clustering groups of similar stories with the visual and semantic information of news story. And then, on the base of the result of the first step, we use different weighted strategies to analyze the news story correlation in a further way, which aim at the different type of news story. The methods can get a better result of the news story correlation analysis by experiments.
D. W. Chen et al., "Combining Multi-Modal Features for News Story Correlation Analysis", Advanced Materials Research, Vols. 268-270, pp. 1040-1045, 2011