Abstract #

Despite the broad range of Machine Learning (ML) algorithms, there are no clear guidelines on how to identify the optimal algorithm and corresponding hyperparameters configurations given an Opinion Mining (OM) problem. In ML, this is known as the Algorithm Selection Problem (ASP). Although Automatic Algorithm Selection or AutoML has proven to be successful in many areas of ASP, it has hardly been explored in OM. This paper explores the benefits of using AutoML in this field. To this end, this work examines to what extent AutoML can be competitive against ad hoc methods (manually select and tune ML pipelines) on Arabic opinion mining modeled from a supervised learning perspective. We compare four state-of-the-art AutoML tools on 10 different popular datasets to human performance. Experimental results show that the AutoML technology can be considered as a powerful approach to support the ML algorithm selection problem in opinion mining.

Citation #

@incollection{Garouani2022,
  doi = {10.1007/978-3-031-02447-4_17},
  url = {https://doi.org/10.1007/978-3-031-02447-4_17},
  year = {2022},
  publisher = {Springer International Publishing},
  pages = {163--171},
  author = {Moncef Garouani and Kasun Zaysa},
  title = {Leveraging the Automated Machine Learning for Arabic Opinion Mining: A Preliminary Study on {AutoML} Tools and Comparison to Human Performance},
  booktitle = {Digital Technologies and Applications}
}