Investigating Unit Weighting and Unit Selection Factors in Thai Multi-document Summarization

摘要

Breaking down documents into small units, unit weighting and unit selection are two important factors in summarization of multiple related documents. This paper presents an investigation on performance of several variants of unit weighting and selection schemes on Thai multi-document summarization. Fifty sets of Thai news articles with their reference summaries are used to evaluate the performance of various weighting and selection methods. Compared to PageRank and Maximal Marginal Relevance (MMR) with four ROUGE measures, the results show that iterative weighting gets higher performance of traditional TF-IDF, the iterative node weighting, query relevance, centroid-based selection, and unit redundancy consideration can help improving summary quality.