Generating Inferences from Qur’anic Verses: A Computational Text Mining Approach
نویسندگان
1 Associate Professor, Computer Science and Engineering Department, Shahid Beheshti University, Tehran, Iran
2 Assistant Professor, Cyberspace Research Institute, Shahid Beheshti University, Tehran, Iran
3 Master's degree in Computational Qur’an Mining, Interdisciplinary Qur’anic Studies Research Institute, Shahid Beheshti University, Tehran, Iran.
4 Assistant professor, Interdisciplinary Qur’anic Studies Research Institute, Shahid Beheshti University, Tehran, Iran
doi
10.37264/JIQS.V4I2.1چکیده
The Recent advancements in deep learning have yielded novel and significant capabilities in natural language processing (NLP) and automatic inference generation. These capabilities are particularly critical due to their resemblance to human reasoning. At the same time, interdisciplinary initiatives have led to substantial advancements in the realms of knowledge and technology. In this study, the Qur’an is examined as a rich source of multiple concepts and teachings. The objective of this research is to employ natural language processing algorithms to derive meaningful and accurate inferences from the English translation of the Qur’an. Inference is defined as the process of deriving a new and logical sentence from two basic and related sentences. The research methodology introduces a model that utilizes transformers and pre-trained language models. Consequently, we construct the set of all unique unordered verse pairs (i<j) from 6,236 verses, totaling 19,440,730 pair evaluations. A fine-tuned BERT-based classifier labels each pair as either exhibiting or not exhibiting a syllogistic relation. Pairs predicted as “Yes” and exceeding a confidence threshold of 0.80 proceed to the subsequent stage, which is inference generation, while all other pairs are discarded. In the following phase, large language models are employed to generate inferences from the selected pairs of verses.
