Abstract—The Qur’an, a sacred text serving as the primary source of guidance for over 1.8 billion Muslims, addresses various life challenges. Ever since 1400 years ago, it has been studied for its literary content and term regularity. Despite many studies of its content from the past to the present, its specialty in terms of the relationship between words is still less studied. Therefore, starting from the hypothesis that the Qur’an has a special regularity in letters, words, and vocabulary, this study aims to utilize the regularity of the relationship between words in the Qur’an for verse classification. This case study can establish the categorization of Islamic literature in the future. Graph mining is proposed to identify the most important words. The graphs of words are constructed using a text corpus of verses in two preprocessing scenarios: with and without stop word removal. This study employs five directed graph centrality measures (degree, betweenness, closeness, harmonic, and eigenvector) as term weighting. The centrality values are then utilized to CNN and RNN for the multilabel classification of Qur’anic verses into eight thematic categories. Our findings reveal that the lowest Hamming loss is achieved by the closeness centrality of a graph without stop word removal using CNN (i.e. 0.1694) and RNN (i.e. 0.1615).
Keywords—qur’an, verse topic, graph mining, neural networks, multilabel classification