<p>How to intervene in a group of agents such that their opinions reach consensus is an important issue in the social sciences. In this paper, we investigate this problem where the opinions of agents evolve according to the weighted median mechanism inspired by the cognitive dissonance theory in psychology. Some agents, referred to as prejudiced agents, are informed of the prejudice, while the other agents, called unprejudiced agents, do not have such information. We provide quantitative results on the minimum number of prejudiced agents needed to make opinions of all agents reach the expected consensus over three classes of proximity-based graphs: <i>k</i>-nearest-neighbor cycle, grid graph, and random geometric graph. In addition to this, we construct the methods to appropriately choose prejudiced agents such that the system reaches consensus on the prejudice over these three graphs. Simulation results are given to verify the effectiveness of theoretical results.</p>