Multi-document Arabic Text Summarization

Abstract

With the rapid growth of data on the internet, there is an essential need for automatic summarization systems. A lot of automatic text summarization researches have been done for English and other languages. Recently, there has been growing interest in the Arabic language by researchers. Many of these researches concerned with single document Arabic text summarization. Multi-document summarization facing many challenges, the main challenges are: redundancy removal, and sentence reordering. In this research, we discussed the possibility of using a single document summarization methods for multi-document summarization, also we proposed a system for multi-document Arabic text summarization based on cross document structure theory. The CST based method help to identify the semantic relationships between sentences across different documents. For redundancy removal we create a novel approach based on splitting the similar sentences into smaller units to eliminate unnecessary ones, and realign the rest of units to form a non-redundant sentence. For evaluation, a Recall-Oriented Understudy for Gisting Evaluation ROUGE evaluation measure is used. The proposed system is applied on news domain using TAC 2011 MultiLing Pilot dataset. The proposed system is compared by ten peer summaries provided from the dataset. The evaluation results show a good performance for CST based method compared to the other peer systems summaries.