1

Although there are appreciable comments and an answer after this question (thank to Robert and Acattle), I am much seeking for a Knowledge branch, (e.g. an "NLP sub-task") than an individual research:

I'm searching for an "NLP subtask" which can help recognizing if a micro-blog post is asking a question and more generally "is answerable" or not. For example the tweet

"A problem shared is a problem halved - so tell us about your one weird phobia"

is not a question but is answerable; or the tweet

"xxx Town is on fire"

is neither a question nor answerable; the tweet

"For our next area spotlight: what's your favorite thing in xxx Garden, and why?"

is a question seeking for answers; but the tweet

"What do you do in your Time Out? xxx and @xxx want to see your x minute films about your London httpx://xxxxx"

though is a question, not seeking for answers.

This recognition, itself, can be of my interest.

Improve this question
3
  • 1
    You need to give a lot more details here about what exactly you want to do, otherwise it will be difficult to suggest something useful. What texts are you concerned with? What is your ultimate aim? And how can a text be a question - do you mean a sentence? – robert May 19 '14 at 16:00
  • Oh yes! I want to check if a micro-blog post, in total, is asking something (is answerable) or not. please tell me if much detail is needed. I want to know which subtask(s) of NLP is much-engaged with this. – hossayni May 20 '14 at 9:53
  • Thanks. Can you edit your original question and add this information, and also give one or two examples of such a micro-blog post? And what are you going to do after identifiying whether it is a question? Provide an answer automatically? – robert May 20 '14 at 10:00
1

I think you might be interested in the paper "Question Identification on Twitter" by Li, Si, Lyu, King, and Chan. This should give you a background on question identification in general as well as an overview of the machine learning techniques typically used for these types of problems. I suspect with this knowledge you should be able to come up with a modified approach that does what you need.

My one caveat is that this type of approach relies on hand-annotated data which is expensive to produce both in terms of time and paying annotators. Unfortunately, the authors of this paper are highly unlikely to share their data with you as the Twitter API license agreement forbids the resyndication of Twitter content. As I said, most modern NLP techniques require large inputs of annotated data and it may not be feasible for you to collect this data yourself.

Improve this answer
1
  • First of all, I thank you so much for your worth answer. more than a research I prefer to have the ability of reviewing the books theorizing an NLP scope. As you know, there are lots of sub-tasks in NLP; co-reference resolution, sentiment analysis, machine translation, automatic summerization, named entity recognition, discourse analysis, morphological segmentation, ... If I know which of them is the most nearest to what I seek for, I can have reviews on some much theorized and much categorized references. – hossayni May 21 '14 at 11:36

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.