Ant Colony System for Multi-Document Summarization

Proceedings of the 27th International Conference on Computational Linguistics, pages 734–744 Santa Fe, New Mexico, USA, August 20-26, 2018. 734 Ant Colony System for Multi-Document Summarization Asma Bader Al-Saleh Department of Computer Science King Saud University absaleh@imamu.edu.sa Mohamed El Bachir Menai Department of Computer Science King Saud University menai@ksu.edu.sa Abstract This paper proposes an extractive multi-document summarization approach based on an ant colony system to optimize the information coverage of summary sentences. The implemented system was evaluated on both English and Arabic versions of the corpus of the Text Analysis Conference 2011 MultiLing Pilot by using ROUGE metrics. The evaluation results are promis- ing in comparison to those of the participating systems. Indeed, our system achieved the best scores based on several ROUGE metrics. 1 Introduction Multi-document summarization (MDS) is a type of summarization in which the contents of a set of doc- uments are represented as a single summary. Today, this type of summarization has become a necessity due to the existence of enormous amounts of information. It reduces the quantity of text by providing a summary that contains the most relevant and important parts. The automatic summarization problem has been studied since the middle of the 20th century. Therefore, the application of several summarization approaches, such as statistical (Radev et al., 2004; Alguliev et al., 2013; Rautray and Balabantaray, 2017) and graph-based (Erkan and Radev, 2004; Shen and Li, 2010; Mosa et al., 2017b) approaches, have been described in the literature. An extractive summary represents a combination of the most important sentences from the source without modifying them. Summary sentences are selected according to the two following approaches. The first is the greedy selection approach, in which the best textual units (e.g., sentences) are selected one item at a time. This approach is widely used in text summarization and is simple and fast; however, it rarely produces the best summaries (Huang et al., 2010). The second approach is the global optimal selection approach, which searches for the best summary rather than for the best sentences. It reduces the summarization task, or at least the step of selecting sentences, to an optimization problem in which the overall score of the output summary is optimized by searching for the best mixture of sentences. In the literature, several summarization objectives have been studied and optimized, such as information coverage, text diversity, and readability. In addition, different meta-heuristics have been applied in the studies to approximate the solution of the summarization problem. One group of such meta-heuristics is swarm intelligence (SI). SI is a nature-inspired population-based type of meta-heuristics that has been applied successfully to the summarization problem (Alguliev et al., 2013; Peyrard and Eckle-Kohler, 2016). Additionally, ant colony optimization (ACO) algorithms have been successfully applied to short text (Mosa et al., 2017a; Mosa et al., 2017b) and single document summarization (Tefrie and Sohn, 2018). Motivated by the importance of the summarization task as well as by the promising results of the studies mentioned above, this paper investigates the application of the …

Content truncated to 500 words