Towards a New Hybrid Approach for Abstractive Summarization

Abstract With the huge amount of Arabic digital data, a summarization system is very helpful to quickly retrieve useful information and save a lot of time and efforts. Two main techniques are used when developing such system: 1) Extractive techniques which consist of returning main sentences based on statistical approaches, 2) abstractive techniques which involve the use of advanced complex processing to generate new sentences representing the summary. The Arabic community has focused mainly on extractive techniques rather than abstractive ones. In this article, we present a new approach for abstractive summarization using conceptual graphs (CGs). We firstly begin with an extractive step to keep only sentences with high weights. Then we generate conceptual graphs for each sentence and proceed to CGs operations including but not limited to: contraction, comparison, join, etc. These operations reduce the number of sentences and words to produce a small summary.