Summarizing Web Documents Using Sequence Labeling with User-Generated Content and Third-Party Sources
説明
This paper presents SoCRFSum, a summary model which integrates user-generated content as comments and third-party sources such as relevant articles of a Web document to generate a high-quality summarization. The summarization was formulated as a sequence labeling problem, which exploits the support of external information to model sentences and comments. After modeling, Conditional Random Fields were adopted for sentence selection. SoCRFSum was validated on a dataset collected from Yahoo News. Promising results indicate that by integrating the user-generated and third-party information, our method obtains improvements of ROUGE-scores over state-of-the-art baselines.