Topic-Sensitive Multi-document Summarization Algorithm

Latent Dirichlet Allocation (LDA), has beenrecently used to automatically generate text corpora topics, and applied to sentences extraction based multi-documentsummarization algorithms. However, not all the estimatedtopics are of equal importance or correspond to genuinethemes of the domain. Some of the topics can be a collection ofirrelevant or background words, or represent insignificantthemes. This paper proposed a topic-sensitive algorithm formulti-document summarization. Our approach is distinguishedfrom existing approaches in that we use LDA model to identifyand distinguish significance topic which is used in sentenceweight calculation. Moreover, beside topic characteristics, thisapproach also considered some statistics characteristics, suchas term frequency, sentence position, sentence length, etc. Thisapproach not only highlights the advantages of statisticscharacteristics, but also cooperated with LDA topic model. Theexperiments showed that the proposed algorithm achievedbetter performance compared the other state-of-the-artalgorithms on DUC2002 corpus.