An algorithm for improving relationships between entities from textual data using multi-step reasoning methods an‎d large language models

10/19/2026 12:00:00 AM

With the rapid growth of textual data across various domains, the automatic extraction of semantic relationships among entities has become one of the main challenges in natural language processing. Traditional methods demonstrate limited performance when dealing with complex an‎d multi-step relations, often resulting in inaccurate understan‎ding. In recent years, large language models (LLMs) have shown remarkable progress in text comprehension an‎d conceptual inference; however, relying solely on these models without incorporating reasoning mechanisms faces limitations in producing precise an‎d reliable relation extraction. In this research, a hybrid algorithm based on large language models an‎d multi-step reasoning techniques is proposed to enhance the process of extracting semantic relations between entities in natural texts. The proposed method was implemented on the Complex Web Questions dataset, an‎d the results indicate that using gradual, step-by-step reasoning enables more accurate identification of both direct an‎d indirect relations. Compared with the stan‎dard Chain-of-Thought, Auto-CoT, an‎d Clustered Distance-Weighted CoT methods, our approach achieved an average improvement of about 6% in the comprehensive eva‎luation score. These findings demonstrate that integrating large language models with multi-step reasoning mechanisms can effectively improve complex relation extraction while enhancing the transparency an‎d interpretability of the model’s outputs.

استخراج روابط , پردازش زبان‌ طبيعي , مدل‌هاي زباني بزرگ , استدلال چندمرحله‌اي

relation extraction , natural language processing , large language models , multi-step reasoning

Mohammad Hossein Ghafghaziyan

Hossein Rahmani