Chinese spelling correction (CSC) based on large language models (LLMs) has achieved remarkable progress in recent years. However, with the rapid evolution of social media platforms, the linguistic context of Chinese has undergone significant changes. Internet slang, characterized by its deviation from conventional spelling norms and its dynamic nature, poses new challenges for traditional CSC systems. Specifically, preserving slang expressions and avoiding overcorrection within these emerging contexts have become critical yet underexplored issues. To address this, we propose CSCIS (Chinese Spelling Corrector for Social-Platform Internet Slang Texts), a novel framework for CSC in Internet slang texts on social platforms. CSCIS introduces a novel multidimensional causal inference designed to preserve Internet slang, and a knockout search algorithm to effectively roll back overcorrections. Extensive experiments demonstrate that CSCIS significantly outperforms existing baselines in slang retention, overcorrection rollback, and overall correction accuracy, while also exhibiting high efficiency.