Scholarlink

This paper presents a hybrid approach that not only retrieves semantically similar English translations of Quranic verse(s) against a query verse but also explains its results by generating natural language based explanations. The work aims to help in a better understanding of Quranic teachings by providing precise and clear explanations emphasizing commonalities in Quranic verses scattered at different places in Quran. The presented approach is a combination of embeddings, ontology and LLMs. Its performance was evaluated against a benchmark dataset, QurSim, and the quality of generated explanations results was evaluated using human evaluators. The hybrid approach performed better than a fine-tuned BERT and a custom-trained Word2Vec. Additionally, experiments show that the explanations generated by LLM(ChatGPT-3.5) are of good quality with a Completeness accuracy of 86% and a Correctness accuracy score of 83% as evaluated by human judges.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

On Combining Embeddings, Ontology and LLM to Retrieve Semantically Similar Quranic Verses and Generate Their Explanations

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年

查看全文
  • Sumaira Saeed,
  • Quratulain Rajput,
  • Sajjad Haider

摘要

This paper presents a hybrid approach that not only retrieves semantically similar English translations of Quranic verse(s) against a query verse but also explains its results by generating natural language based explanations. The work aims to help in a better understanding of Quranic teachings by providing precise and clear explanations emphasizing commonalities in Quranic verses scattered at different places in Quran. The presented approach is a combination of embeddings, ontology and LLMs. Its performance was evaluated against a benchmark dataset, QurSim, and the quality of generated explanations results was evaluated using human evaluators. The hybrid approach performed better than a fine-tuned BERT and a custom-trained Word2Vec. Additionally, experiments show that the explanations generated by LLM(ChatGPT-3.5) are of good quality with a Completeness accuracy of 86% and a Correctness accuracy score of 83% as evaluated by human judges.