A comparative study on collectives of term weighting methods for extractive presentation speech summarization

Abstract

This paper presents a comparative study of collectives of term weighting methods for extractive speech summarization of Mandarin Presentation Speech. The summarization process can be considered as a binary classification process. The collectives of different term weighting methods can provide better summarization performance than each of them with the same classification algorithm. Several different unsupervised and supervised term weighting methods and their collectives were evaluated with summarizer based on support vector machine (SVM) classifier. The majority vote strategy is used for handling the collectives. We show that the best result is provided with the vote of the collective of all term weighting methods. We also show that Term Relevance Ratio (TRR) gives more contribution for presentation speech summarization than other term weighting methods.