Automatic Domain-Specific Text Summarisation With Deep Learning Approaches
Hellesoe, Lui Joseph
MetadataShow full metadata
Text summarisation has been recognised as a critical Natural Language Processing task, attracting significant attention from researchers and practitioners. Various domains have widely adopted it. For example, text summarisation of news, articles and book chapters can produce a short text, assisting the readers with grasping the main idea rapidly. In the medical domain, practitioners apply it to summarise their questions, and in the legal domain, practitioners also use it for summarising Judges decisions. However, it is challenging to control the summarised output by producing domainspecific summaries since the focus of domain-specific information may be ignored. In recent years, automatic text summarisation has become a hot research topic. This thesis proposes a novel approach for domain-specific document automatic text summarisation. With this, users can realise more efficient reading and understanding of the main contents of a document after the summarisation. In order to solve the problem of domain-specific summarisation, we propose a hybrid model that combines three kinds of embedding approaches: domain, focus and context embeddings. We apply the proposed approach to the MeQSum and LegalCosts datasets for evaluating the performance and effectiveness of using hybrid embeddings for specialised document summarisation. The experimental results demonstrate that our model outperforms state-of-the-art algorithms in automation and summary quality.