I am a computer engineer working with text (NLP). As part of my PhD, I am conducting some experiments on automatic summarization, i.e., automatically generating a summary / abstract of a text. One of the datasets I am using contains patents' text. The dataset idea is to generate the patent Abstract given the Detailed description / Detailed description of the embodiment. For example, for this patent, one has to automatically generate the Abstract given the Detailed Description.
Since this task seems pretty hard to me even for humans (certainly it would be for me, but my knowledge of patents is very limited), my questions are:
- Given a patent text (no figures), is the Detailed description enough for an expert to write the Abstract?
- What other sections might be important/useful/easier to use?
- And how is the process carried out, in reality, when writing a patent?