In natural language processing, word-sense disambiguation (WSD) is an open problem concerned with identifying the correct sense of words in a particular context. To address this problem, we introduce a novel knowledge-based WSD system. We suggest the adoption of two methods in our system. First, we suggest a novel method to encode the word vector representation by considering the graphical semantic relationships from the lexical knowledge-base. Second, we propose a method for extracting the contextual words from the text for analyzing an ambiguous word based on the similarity of word vector representations. To validate the effectiveness of our WSD system, we conducted experiments on the five benchmark English WSD corpora (Senseval-02, Senseval-03, SemEval-07, SemEval-13, and SemEval-15). The obtained results demonstrated that the suggested methods significantly enhanced the WSD performance. Furthermore, our system outperformed the existing knowledge-based WSD systems and showed a performance comparable to that of the state-of-the-art supervised WSD systems.