In order to overcome the data sparseness problem of traditional Word Sense Disambiguation (WSD) methods, a new WSD method based on knowledge context was proposed. The method is based on the assumption that sentences within one article share some common topics. Fisrt, similarity algorithm was used to obtain sentences with the same ambiguous words in the article, and those sentences could be appropriate knowledge context for ambiguous sentences and provided disambiguation knowledge. Then a graph-based ranking algorithm was used for WSD. The experimental results of real data show that, when there are two knowledge context sentences and the window size is 1, the disambiguation accuracy of this method is increased by 3.2% compared to the baseline method (OrigDisam).