Sentence Scoring in Multi-document Summarizing under Topic Model LDA

摘要

In this paper automatic multi-document summarizing in a greedy framework is studied, where sentences are selected based on their contribution for the theme construction of the summary. The scores of sentences are evaluated based on their topic representations obtained from LDA (Latent Dirichlet Allocation), which is a probabilistic topic model. Consistent probabilistic representations of the relations between texts and topics are first proposed, and then two scoring methods are developed based on these representations. In addition the sentence length as an important factor in document summarizing is also studied. Experimental results show the pertinence of these probabilities, and the effectiveness of our scoring methods.