In information retrieval, the lexical mismatch is a common problem where texts in documents and user queries have different forms even though they have the same meaning. The problem occurs more frequently when people search for jargon words because they tend to use descriptive queries and the jargon words interchangeably such as “hypertension” and “high blood pressure”. Query reformulation (QR) is one of the techniques that overcome the lexical gaps by transforming users’ descriptive queries into appropriate jargon queries. This study proposes a novel graph-based QR system that uses a dictionary, where the model does not require manually annotated data. Given a descriptive query, our system predicts the corresponding jargon word over a graph consisting of headwords and words in their definition. We use a graph neural network to represent the relational properties between words and the curriculum learning to train the networks stably and gradually adjust to hard samples. Moreover, we propose a graph search model that finds the target node in real-time using the relevance scores of neighborhood nodes. By adding this fast graph search model to the front of the proposed system, we reduce the reformulating time significantly. Experimental results on two datasets show that the proposed method can effectively reformulate descriptive queries to corresponding jargon words as well as improve retrieval performance under several search frameworks.