Machine Reading Comprehension using NLP

Machine Reading Comprehension using NLP

February 7, 2019 DATAcated Challenge 0

Machine Reading Comprehension(MRC) is using NLP  for text comprehension. That is answering questions in context of a text passage.
MRC has great potential to augment human capabilities and below are just few of the potential use cases:

1. Better Online Search – Current search only curates the documents or web results based on keywords. Actual reading and comprehending is performed by people. If machine is able to comprehend the documents, the search experience can be revolutionized.

2. Knowledge Management in an Organization – Any big organization struggles in knowledge management of its various systems and depends on domain experts for system maintenance. A lot of this knowledge exists in the form of documents which gets lost in cold storage of computer hard drives with time. A machine comprehending this knowledge and presenting solutions for technical issues or aiding in knowledge can greatly augment the human capabilities.

3. Chat bot based solution for product information – Imagine you buy a new machine or software. And are required to do self-assembly, installation or configuration. You have to go through time consuming process of reading hundreds of pages of product manual. Instead, if a chat bot based MRC system is implemented, one need not go through the document manually, and just ask the Chat bot for relevant information.

Current Challenges for MRC – There is a lot of research in this area and if we assume that the answer is a continuous span in the context paragraph, model can be trained to output the starting point and end point of text with great accuracy.

But in real world, best answer may require analysis and reasoning over multiple snippets in the paragraph which is a challenge to be solved to create great MCR solutions.

The future has limitless possibilities.

* Source – Taken ideas from KDNuggets article.


By: Abdul Rehman


Leave a Reply

Your email address will not be published. Required fields are marked *