Comparative Study of Recurrent and Dense Neural Networks for Classifying Maritime Terms

Despite its importance, the extraction of domain-specific terms has not been sufficiently studied. This paper proposes an automated approach for semi-supervised term extraction from Greek-language legal documents related to the shipping/maritime industry. The approach employs a deep learning scheme based on a machine learning model that uses several linguistic features and word embeddings, and is trained using two different kinds of ground truth: the size of words in characters and the use of freely available nautical dictionaries. In addition, the approach was able to extract a large number of domain-specific terms that were not included in existing dictionaries. The results show that the proposed approach outperforms human annotation when a semi-supervised method is used for ground truth. These results suggest that the proposed approach could be a useful tool for term extraction from legal documents in the shipping/maritime industry, and could potentially be adapted for other domains and languages as well.