<p>This paper introduces Ct-K, a novel approach for discourse anaphora resolution that integrates Centering Theory with K-means clustering. The proposed method enhances discourse coherence modeling by combining linguistic insights with scalable machine learning techniques. Ct-K identifies and clusters discourse entities while preserving semantic relationships and provides a robust framework for analyzing multilingual and domain-diverse text. Experimental evaluations for 200 documents demonstrate its superiority over traditional methods in precision, recall, and F1 score. This study highlights the potential of Ct-K to improve clustering quality and coherence in natural language processing tasks, and gives a foundation for future advancements in discourse analysis.</p>