Online interpersonal harm, such as harassment and discrimination, is prevalent on social media platforms. Most platforms adopt content moderation as the primary solution, relying on measures like bans and content removal. These measures follow principles of punitive justice, which holds that perpetrators of harm should receive punishment in proportion to the offense. However, these strategies often fall short of addressing the needs of affected individuals — the survivors — who are typically excluded from decision-making and left with various unmet needs. My dissertation adopts a restorative justice lens and investigates ways to empower online harm survivors in addressing their unique needs. This approach emphasizes survivors' needs and agency, reconceptualizes views on perpetrator accountability, and mobilizes community resources for a collective response to the issue. Through interviews and co-design sessions with survivors, I identified key survivor needs such as sensemaking, emotional support, safety, retribution, and transformation. Building on these insights, I focused on survivors' needs for sensemaking and developed a social computing system to facilitate a structured sensemaking process, connecting survivors with available resources and stakeholders in addressing the harm. Furthermore, I applied a restorative justice lens to understand how survivors, perpetrators, and moderators currently navigate harm within existing moderation practices, examining both the opportunities and challenges of integrating restorative justice practices into the content moderation landscape. My research highlights the urgent need for social media platforms to incorporate justice and ethical values into their operational frameworks. It advocates for a survivor-centered approach to addressing online interpersonal harm, viewing it as a multi-stakeholder, cross-platform process. This approach calls for a shift towards viewing harm as a communal challenge and emphasizes cultivating a resilient community culture in the long term, rather than perceiving harm as isolated incidents within a perpetrator-centric framework.