Published Date Extraction System A semi-supervised approach of extraction

摘要

The need to extract a meaningful or relevant dates like published date from an unstructured document is a very vital cog in the wheel of information extraction and data mining field. The current approaches usage DOM (Document Object Model) manipulation for an HTML document or regex expression and rules from metadata which are not so accurate for different types of publication. The recent work in this area mainly focused on web pages and HTML pages with some good accuracy. Our approach took a leaf from those works for HTML, and along with that it extensively covers PDF document, Blog articles, and Websites. It supports several types of documents like News Articles, Patents, Scientific Articles/Journal in PDF format, Blogs, Websites and more. It also has the capabilities to learn over the period and feed the learnings back to the system as trained model. Our algorithm comprises of both supervised and unsupervised steps, and it uses natural language processing techniques.