Cascaded Regression Analysis Based Temporal Multi-document Summarization

Temporal multi-document summarization (TMDS) aims to capture evolving information of a single topic over time and produce a summary delivering the main information content. This paper presents a cascaded regression analysis based macro-micro importance discriminative model for the content selection of TMDS, which mines the temporal characteristics at different levels of topical detail in order to provide the cue for extracting the important content. Temporally evolving data can be treated as dynamic objects that have changing content over time. Firstly, we extract important time points with macro importance discriminative model, then extract important sentences in these time points with micro importance discriminative model. Macro and micro importance discriminative models are combined to form a cascaded regression analysis approach. The summary is made up of the important sentences evolving over time. Experiments on five Chinese datasets demonstrate the encouraging performance of the proposed approach, but the problem is far from solved.