I've been working on document level sentiment analysis over the last year. Document level sentiment analysis provides the sentiment of the complete document. For example - The text "Nokia is good but vodafone sucks big time" would have a negative polarity associated with it as it would be agnostic to the entities Nokia and Vodafone.

How would it be possible to get entity level sentiment, like positive for Nokia but negative for Vodafone? Are there any research papers providing a solution to such problems ?

  • could you expand your thoughts on the problem? give some parallel example, maybe? – Tames Jun 21 '12 at 15:48
  • @Tames a sample text could contain multiple entities like organization, people, services, etc., so instead of having a single sentiment value for the complete text, how can we get sentiment value related to each entity? Another example would be - "I love Microsoft but Apple sucks", here we would like to detect 2 sentiment values. One related to Microsoft(positive) and another related to Apple (negative). – Jasneet Jun 21 '12 at 18:17
  • You have to at least (1) recognize entities (2) distinguish positive, neutral, and negative polarities (3) parse to at least the immediate constituent level, preferably with independent phonotactic constituents overlapped. Then you can get probability values. Strong sentiments tend to be associated with entities that are mentioned in the same syntactic and phonotactic constituents; the closer the better. But at least major constituents have to be recognized. – jlawler Jun 21 '12 at 18:24
  • 1
    @jlawler it seems that you have almost answered the question in you comments. Why don't you turn them into a real answer, perhaps elaborating those ideas a bit further? – Otavio Macedo Jun 21 '12 at 20:17
  • 1
    Some of the people in my lab are currently developing a Korean Langauge version of MPQA. They found it necessary to create an entity level unit, as you describe, in addition to MPQA's sentence level sentiment tagging. They call it a "SEED tag". They haven't published any of their work yet but they will be giving a presentation at PACLIC 26. As they've explained it to me, no existing corpus does this type of entity level sentiment tagging. I really wish I had some papers I could show you. – acattle Jun 22 '12 at 6:21

The fundamental method for sentiment analysis relies on precompiled polarity lexicons. (This in itself is questionable because word meaning is heavily context-dependent but it provides good approximate results).

Refining it to phrase-level opinions simply involves combining such a lexicon with a sentence parser. There is indeed a lot of research going on on this subject, mainly because it's well funded. Indeed sentiment analysis is one of the areas of NLP where business application is straightforward, since our corporate world loves to know what consumers think of it.

For a reference paper, you can have a look at [McDonald, Hannan, Neylon, Wells and Reynar 2007 - Structured models for fine-to-coarse sentiment analysis]

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.