Businesses need to adhere to certain regulations to remain compliant. They want to expand or move to a new geography and find itself subject to a slightly different set of regulations. Also, regulations themselves change over time and force the business to change its internal working to remain compliant. When a compliance officer is presented with a new regulatory document, he has to manually compare corresponding sentences between previous and the new version. While most studies in text mining have focused on measuring textual similarity, textual entailment detection and paraphrase identification etc., there has been very little focus on the problem of change tracking (CT). Change tracking can be defined as the task of identifying the phrase pair(s) that captures the semantic difference between two given sentences, and plays an important role in domains such as financial regulatory compliance where core changes introduced by regulators to existing regulations need to identified quickly. Naturally, the change tracking has to satisfy the minimality and comprehen-siveness requirements even in presence of complex language structure, context dependence and paraphrasing between com-pared sentences. In this paper, we address these challenges and devise a graph-based approach called DeepAntara1 and show its performance for change tracking task over multiple sentence pairs extracted from different versions of publicly available financial CRS treaties.