I have millions of lines of statements containing both subjective (like I prefer the red skirt) and objective (Washington was born on February 22, 1732) statements or opinions. How can I separate them? Not manually.

By "objective", I mean if the object or predicative of the sentence changed, it would be in conflict with the truth. For illustration, in "She prefer red skirt.", if we change "red" or "skirt", the (new) statement would remain right, which can not apply to statements like "Washington was born on February 22, 2016".

The statements which should be pushed out:

  1. Washington was born on February 22, 1732
  2. Washington was born on February 21, 1731
  3. Obama was born on February 22, 1732
  4. Nobody A was born on February 22, 2016
  5. Red is blue
  6. Color is a day
  7. I'm a robot
  8. You are a person

Statements which should be kept:

  1. I like coffee
  2. Red is my favorite color
  3. I hate him
  4. You are awesome
  5. You are clever
  6. I was born years ago
  7. You are very old
  8. Tomorrow will be better
  • $\begingroup$ Please give lots more examples of what you would label subjective and objective. Can we get away with just looking at the main verb, e.g. 'think/feel/believe/prefer/...' or noun/adverb 'in my opinion/preference/belief'? Show us ones that fail a simple keyword filter like that. $\endgroup$ – smci Dec 12 '16 at 22:48
  • $\begingroup$ You're right. The head word is important. It might be tackled by extracting the head word or all verbs in the sentence and then classify it using a discriminative model, for instance logistic regression or SVM. $\endgroup$ – Lerner Zhang Dec 13 '16 at 2:14
  • $\begingroup$ Nouns or other words may also be relevant. $\endgroup$ – Lerner Zhang Dec 13 '16 at 2:17
  • $\begingroup$ We still need lots more examples of what you would label subjective and objective, and you suggest below you might add a third class: neutral. $\endgroup$ – smci Dec 13 '16 at 3:22
  • 1
    $\begingroup$ You'll never get an answer on an "objective" classification of objective vs. subjective. If you still want to answer, perhaps you can add details of how such a model will be used, and then the right binary classification can be decided then. $\endgroup$ – Paul Dec 16 '16 at 13:21

There isn't really an "objective" statement.

In your example: what if he wasn't born that day? Same statement, but is it still objective?

You may assume that "1 + 1 = 2" is an objective statement. But what if I'm doing binary math, and 1+1=0 then? So even that is subjective.

How would a machine tell apart these things, where philosophers will disagree with each other?

All you can do is provide training data examples of "your" (subjective) idea of objectiveness.

  • $\begingroup$ It doesn't matter either the statement is wrong or right. I should avoid a QA system providing any question with an absolute answer which increases error rate using the generative NN system. $\endgroup$ – Lerner Zhang Oct 14 '16 at 1:09
  • 2
    $\begingroup$ A better counter example would be "she prefers the red skirt" Is this an objective or subjective statement? $\endgroup$ – Diego Oct 23 '16 at 3:45
  • $\begingroup$ I would say it can be viewed as a subjective statement since "prefers" indicates a feeling(which would be changing and not fixed). $\endgroup$ – Lerner Zhang Dec 13 '16 at 1:57
  • $\begingroup$ Maybe it would be seemed as a neutral one? $\endgroup$ – Lerner Zhang Dec 13 '16 at 2:20
  • 2
    $\begingroup$ Objectively neutral, or subjectively? :-) "She is hungry." Objective? Subjective? As you can see, we don't agree on what is objective - so the term "objective" is subjective... $\endgroup$ – Has QUIT--Anony-Mousse Dec 13 '16 at 2:27

I was looking for some sentiment datasets when I encountered a so-called subjectivity analysis on this page. I thought what I meant can be found in this paper.


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.