Understanding the influence of hyperparameters on text embeddings for text classification tasks

Abstract

Many applications in the natural language processing domain require the tuning of machine learning algorithms, which involves adaptation of hyperparameters. We perform experiments by systematically varying hyperparameter settings of text embedding algorithms to obtain insights about the influence and interrelation of hyperparameters on the model performance on a text classification task using text embedding features. For some parameters (e.g., size of the context window) we could not find an influence on the accuracy while others (e.g., dimensionality of the embeddings) strongly influence the results, but have a range where the results are nearly optimal. These insights are beneficial to researchers and practitioners in order to find sensible hyperparameter configurations for research projects based on text embeddings. This reduces the parameter search space and the amount of (manual and automatic) optimization time.