AI Classification of Linguistic Expression Between The Quran, The Hadith, and Pre-Islamic Poems Using an LSTM Deep Learning Model

Abstract

The Quran is God's (Allah subhanah) universal and final message to humanity through his prophet Muhammad peace and blessings be upon him (PBUH). This research aims to prove the distinct and separate nature of the linguistic style of expression used in the Quran vs. the Hadith vs. Pre-Islamic Poetry in Mecca during the prophet's era, known as the "Ten Hanging Poems". While all these sources were all in Arabic language, we demonstrate that they are each distinct in their style of expression and belong to separate authoritative sources. The prophet's (PBUH) style of natural language expression is transmitted verbatim through his Hadith via each Hadith's trusted chain of narrations. While the Quran is authored by Allah subhanah and transmitted to his prophet Muhammad PBUH through the angle Jibreel (Gabrael) PBUH. This research will demonstrate the distinct and separate source of authorship between the Quran, Hadith, and Pre-Islamic Poetry using an Artificial Intelligence deep learning model of Long Short-Term Memory (LSTM) network. The Quran has 6236 verses, and we extracted Prophet's words (PBUH) through 5181 Hadiths from Sahih Bukhari's book with verified trusted chain of narrations, as well as used 858 lines of Pre-Islamic Poetry. For the Hadiths processed, we purposefully avoided text not expressed by the prophet, including narration chain, expression by others, or quotes of Quran verses.. We trained three different models Net21, Net20, Net19 on 25%, 20%, and 15%, respectively, of total 6236 Quran verses with randomized order of the verses so as to avoid bias of model due to verse length. Similarly we trained the three models on the same percentage of available text from the Hadiths, and Poems, and then tested each of the models on the residual 75%, 80%, and 85%, respectively, of Quran verses, Hadiths, and Poems. Accurate classification of the three LSTM Models of testing Quran verses was 98.58%, 98.95%, and 83.47% respectively. Accurate Hadith's classification accuracy of the three LSTM Models 98.97%, 99.73%, and 99.59% respectively. Accurate classification of the three LSTM Models for the Poems was 100%, 100%, 100% respectively. These results demonstrate the distinct nature of the expression style of authorship between the Quran, the Hadith, and the Pre-Islamic Poetry leading to the conclusion that they are indeed from different sources of authorship. This research results provides objective scientific proof that the Quran is not the creation of the Prophet Muhammad (PBUH) but is from a divine source (Allah subhanah). It also demonstrates that the Prophet's style of expression in his speech in the verified trusted Hadith was not influenced by the Quran. It further demonstrates that the Prophet's style of expression was not influenced by the Poetry style that was common in his era in Mecca where he grew. Finally, this research demonstrates that the Quran did not follow the poetic style of expression common in the era of Pre-Islamic Mecca, but rather was distinguished in its own class of expression style that fascinated and attracted people to it as it was different from what they heard before in poetry. These facts are also mentioned in the following 7 Quran verses that the Quran is not the creation of the prophet Muhammad (PBUH) as falsely claimed by some people. It is also mentioned that the Prophet was "not a poet" in 3 Quran verses.

Deep learning LSTM network models are suitable for this application of text comparison because of their ability to hold memory states of text sequences and because they adapt well extracting key features for classification via its state memory structure even in the presence of scarcity of input training data. In this paper we will overview the structural properties of LSTM deep learning models, and the pre-processing of Arabic language text into a numeric form accepted by the model for training and testing, and finally we will demonstrate the results of the model application and review conclusions.