This topic obtains the data of featured vocabulary under the technical architecture of big data platform and saves it in the form of dataset. Standing on the perspective of the principle of translation of featured words in foreign propaganda, the improved K-means algorithm and attention mechanism are utilized to design the translation model of featured words. The model of this paper is validated and analyzed from two aspects, namely, performance indexes and application effect, respectively. In the six performance indexes, this paper’s model performs better compared to the other two control models. After the experience, the control group and the experimental group show a significant difference, i.e., the introduction of data mining algorithm is more effective in translating the featured vocabulary on the traditional model.