2

I have to identify explicit and implicit aspect from a movie review.

The explicit aspect is the one that expressed in a review directly such as the price of the camera is reasonable. Here, the aspect is price and the opinion word is reasonable which are mentioned directly here. However, when someone says when I watched this film, I Hope it ended as soon as possible, here the writer means that the movie is very boring. However, there is neither opinion word nor aspect word is mentioned here. we considered here that the sentence hold feature-opinion pair movie-boring.

Now, the question is that how we can identify implicit aspects from unstructured text? While searching I have found that dependency parser will help to make a tree structure of sentence. The next question is which classifiers should be used to measure the accuracy of our system by using dependency parser?

jbsu32
  • 1,036
  • 1
  • 11
  • 31
Jibran Mir
  • 1
  • 7
  • The simplest method I can think of is: Make the task formal - choose one or more of the established text mining / information extraction methods (e.g. text classification, (named) entity extraction, event extraction, relation extraction). Manually create gold standard data (collect some texts and add annotations to them) and then use machine learning or similar methods that are capable to fulfill these tasks... – dedek Sep 07 '16 at 06:30
  • I think you have a long way ahead to go... – dedek Sep 07 '16 at 06:34
  • I know this process and i am worried about implicit aspect identification. Explicit aspects can be identified using NER and CRF classifier. but implicit aspect identifier is the real issue for me – Jibran Mir Sep 08 '16 at 08:00
  • Try to develop some good learning features for implicit aspect identification... But it will all end up as usual: precision, recall - how good depends on the task difficulty, data size and consistency of human created annotations. – dedek Sep 08 '16 at 08:46
  • 2
    I don't recommend anything like "try make computer to understand the actual meaning of the text". We are not that far yet. But still, if you have some briliant idea Jibran Mir, you may be the first successful one ;-) – dedek Sep 08 '16 at 08:49
  • 1
    I understand your reservations however, we are researchers this is our job to find new ways. – Jibran Mir Sep 08 '16 at 15:56
  • 2
    I dont believe in brilliant ideas, I believe in doing something or nothing. – Jibran Mir Sep 08 '16 at 16:01

0 Answers0