2
$\begingroup$

I have to read a lot of papers, and I thought that I can use an A.I. to read them and summarize them. Maybe find one that can understand what the papers are talking about it seems a lot to ask.

I think I can use natural language processing. Is it the right choice?

I'm sorry, but I'm new in A.I. and I don't know much about it.

$\endgroup$
  • $\begingroup$ I don’t hunk there exists such an AI that is open. Yet. Might be wrong though, so just hang in there! $\endgroup$ – Andreas Storvik Strauman May 27 '18 at 17:24
  • $\begingroup$ Besides searching to something done, I'm also asking, how can I do it? Maybe, I can use natural language processing. $\endgroup$ – VansFannel May 28 '18 at 7:38
  • $\begingroup$ I am also interested to know if there is any online service or API that does this. For example, it scans documents and answers the questions as per the facts it learnt from the docs. From my understanding, Google Cloud NL can analyse documents to extract information but I don't know if it can also answer your questions. With Google you can have docs in Q/A and it parses intents from the docs but again not sure if it is answering your questions? Do we need another service to feed the analysed docs to chat with AI? $\endgroup$ – xbmono May 30 '18 at 6:46
  • 1
    $\begingroup$ This field of AI is called 'automatic text summarisation'. An Internet search of that term will get you started. $\endgroup$ – DrMcCleod May 30 '18 at 7:40
2
$\begingroup$

When automating tasks involving text, NLP techniques are definitely the way to go! Let me be frank, though, starting from scratch and "read and understand" will be harder with today's tools than directly reading the papers.

Some typical required skills are below, and the bar height will depend on your mileage:

  • Knowledge of NLP (e.g. terminology like embeddings, bag of words, even perhaps tf-idf).
  • Knowledge of NLP libraries and implementation frameworks (e.g. Stanford's or TensorFlow).
  • Knowledge of a programming language (e.g. Python or Java).

In short, the goal you set will probably require many more papers than expected. But if you succeed only the automated reading / summary part, you have a bright future ahead (let aside understanding, a way more contentious problem).

Just for the sake of putting an example, companies like Iris.ai and others try to solve related problems---great potential, but definitely challenging.

$\endgroup$
  • $\begingroup$ I do not usually understand the papers I read, so I'm looking for some help (AI). Thanks for your answer. $\endgroup$ – VansFannel May 30 '18 at 6:49
  • $\begingroup$ Sorry just for my own understanding, are you saying with the frameworks like Stanford's and programming language we can write an application that reads and analyses documents and answers questions or this is still a challenging issue? How does these chat bots like Oracle's do this? $\endgroup$ – xbmono May 30 '18 at 6:57
  • 1
    $\begingroup$ @xbmono The existing tools and techniques have qualities and limitations, yet they can be used for many problems that have not been tried yet. For example, one of my previous employers used the Stanford NLP library to perform automated comment moderation (similar to spam filtering). Now the tools have limitations, and there are a lot of challenges. NLP libraries usually break facing indirect speech, some forms of humour, etc. And NLP libraries are usually ignorant of context, e.g. in a discussion thread. A lot possible, and a lot yet to do! $\endgroup$ – Eric Platon May 30 '18 at 8:38
  • 1
    $\begingroup$ @EricPlaton Thanks, but it was only a joke. Thanks again. $\endgroup$ – VansFannel May 30 '18 at 8:45
  • 1
    $\begingroup$ I have received this on email a few minutes ago, coincidence?: software.seek.intel.com/DLNaturalLanguageWebinar_Reg $\endgroup$ – VansFannel May 30 '18 at 17:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.