Conversations on social media platforms are dynamic, mutating over time as both the participants and opinions change. If a method were created that could reliably quantify these changes, it could be used as an analysis metric or as a training input for neural networks. In this paper, the various techniques used by linguists and computer scientists are evaluated, eventually finding a suitable approach rooted in semantic analysis. In this approach, each sentence in a conversation is parsed into a dependency tree, from which true or false propositions can be extracted. The propositions are then compared against one another to find disagreements and consensus between the conversation’s participants. Using this approach, a small case study is conducted on Twitter threads to evaluate its efficiency and reliability. Ultimately, the current implementation is found to be unreliable due to difficulties involving contextual information, sarcasm, and word-sense disambiguation. Solutions to these problems are discussed, and it is likely that they will be overcome with more research.