Whenever the artificial intelligence is used, it doesn’t deliver the 100% accuracy by its nature. As some content is not precisely defined by the nature of its file format and by its technical capabilities, some inaccuracies in minority of cases are expected in:
- Web articles extraction,
- PDF content extraction,
- Detection of the text in images (aside from image files / camera import in some cases it may be related to the PDF files which may contain only photographs of the pages, or even worse combination of pages with text and pages with photographs of texts where it becomes hard for the app to understand what is intended purpose),
- Detection of citation in the text.
On the large diversified sample all of those technologies should work well and produce proper results in significantly more than 90% of cases. However if you are using non-diversifed sample (like from one publisher) you may register 100% inaccuracy in such a sample. As those technologies are in the continuous development samples that don’t work are always welcome and they frequently lead to increased accuracy in the future versions of the app.