Abstract: Reinforcement learning is applied in a multi-agent environment to enable effective determination of user intent classification from documents (e.g., chat, emails or another mode of communication by a user). Although different agents may implement different learning algorithms, they communicate with each other to learn and adjust their behavior by observing peer agents. Using a reinforcement learning (RL) framework, a method integrates each agent's prediction of user intent, as a sequence of tokens in the document are being analyzed. The method continues to refine its observation until it reaches the end of the document. This approach is more effective in uncovering refined linguistic features of words in the document, when read sequentially from start to end.