Text summarization approaches could be broadly classified into two categories: extractive and abstractive. Extractive approaches aim to select the most important pieces of information from an original document without adding any external material to the generated summary or having any deep understanding of the language. Abstractive approaches require a deep understanding of the language; and we find just few work in this direction since it aims to create a shorter version of the original document but not restricted to the material present in the original document. Most of the approaches that have followed an abstractive paradigm rely on predefined templates and cannot be imported to the open domain. So, my question is, do you think that it is possible to propose in the near future approaches that could deal with abstractive text summarization in the open domain? Or maybe using templates is the best choice?
Abstractive approaches require deep knowledge of the target domain and complex substitution rules that require intensive groundwork and corpus specific tagging. You would need extensive semantic and pragmatic information to carry out this type of processing. Due to the domain knowledge requirements I would propose first a template approach from which to study if generalities in procedures can be obtained and then take it from there to see if open domain alternatives are possible.
Just a thought
I think it will be difficult. Since, it needs some supervised learning techniques in the specific domain in which the abstract created from it using the domain knowledge
Yes it will be complex. because abstractive methods will be based on supervised learning, and to achieve it one domain will be considered at a time...
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence