SemG-TS: Abstractive Arabic Text Summarization Using Semantic Graph Embedding

This study proposes a novel semantic graph embedding-based abstractive text summarization technique for the Arabic language, namely SemG-TS. SemG-TS employs a deep neural network to produce the abstractive summary. A set of experiments were conducted to evaluate the performance of SemG-TS and to compare the results to those of a popular baseline word embedding technique called word2vec. A new dataset was collected for the experiments. Two evaluation methodologies were followed in the experiments: automatic and human evaluations. The Rouge evaluation measure was used for the automatic evaluation, while for the human evaluation, Arabic native speakers were tasked to evaluate the relevancy, similarity, readability, and overall satisfaction of the generated summaries. The obtained results prove the superiority of SemG-TS.