As regulators escalate their scrutiny of earnings management practices, the effectiveness of traditional methods in detecting earnings management is often criticized due to the complexity of financial statements. This research proposes a textual comparative earnings management (TCEM) metric by leveraging the information disparity between social media textual data and analyst reports with the help of natural language processing (NLP) techniques to identify accrual earnings management practices. By highlighting the inadequacy to solely relying on analyst reports, we aim to show that TCEM could serve as an effective detector in detecting earnings management. The research demonstrates the value of social media posts as alternative data to detect such financial malpractices, complementing the shortcoming that analysts might conceal negative information, thereby addressing the information asymmetry from the theory of information suppression. Moreover, our detector may serve as a new regulatory technology (RegTech), aiding regulatory agencies to prevent potential financial risk.