Analysis of the impact of parameters in TextGCN

Deep learning models uses many parameters to work properly. Asthey become more complex, the authors of these novel models cannotexplore in their papers the variation of each parameter of theirmodel. Therefore, this work describes an analysis of the impact offour different parameters (Early Stopping, Learning Rate, Dropout,and Hidden 1) in the TextGCN Model. This evaluation used fourdatasets considered in the original TextGCN publication, obtainingas a side-effect small improvements in the results of three of them.The most relevant conclusion is that these parameters influence theconvergence and accuracy, although they individually do not constitutestrong support when aiming to improve the model’s resultsreported as the state-of-the-art.