Utilizing User Posts to Enrich Web Document Summarization with Matrix Co-factorization

In the context of social media, users tend to post relevant information corresponding to an event mentioned in a Web document. This paper presents a model to capture the nature of the relationships between sentences and user posts such as relevant comments in sharing hidden topics for enriching summarization. Unlike the previous methods which usually base on hand-crafted features, our approach ranks sentences and comments based on their importance affecting the topics. The sentence-comment relation is formulated in a share topic matrix, which presents their mutual reinforcement support. Our newly proposed matrix co-factorization algorithm computes the score of each sentence and comment and extracts top m ranked sentences and m comments as the summarization. Experimental results on two datasets in English and Vietnamese of the social context summarization task and DUC 2004 confirm the efficiency of our model in summarizing Web documents.