Online social networks have transformed communication, serving as key platforms for sharing and consuming information. These networks expose users to a range of opinions, either intentionally or incidentally. However, recommendation systems often favor similar content, overshadowing diverse, niche, or novel perspectives. This bias exacerbates challenges such as fake news, filter bubbles, and opinion polarization. This paper introduces a framework to promote diversity in content-based social networks by framing information exposure diversity as an optimization problem. We focus on locally modifying the user-content graph by adding edges to maximize a diversity metric from an individual users’ perspective. Importantly, we define diversity for two semantics: user-user and user-item recommendations. We formalize the concept of information exposure, linking it to established models in the literature, and propose several algorithms to address this problem, including gradient descent-based and greedy methods. Experiments on various real-world datasets show that our algorithms are better than state-of-the-art methods in achieving higher diversity.