5

What is difference between sentiments (positive and negative) and emotions in text mining (NLP)? For example Anger is negative emotion as well as negative sentiment both seems the same. Vijay Nadadur, Creator of SentiRank, an algorithm which ranks sentiment in a text, Suggest Bio Sentiments can be expressed mostly in binary format (+ve & -ve) in a simplistic way. To add further, you may varying degrees of +ve and -ve sentiments, and perhaps neutral. However, emotions have multi-dimensions. Anger for sure is -ve sentiment, and so is sadness, but they aren't really the same.

To talk more specifically about text mining (NLP based), it's much simpler to do sentiment analysis but very hard to carry out emotional analysis. The next level of sentiment analysis is the field of intent analysis where few researchers have been working on mining out intent from the chunk of text, which seems of very high business value.

Adil
  • 105
  • 1
  • 4
  • 10

2 Answers2

6

One could view emotion to be a finer-granularity sentiment. As you mention in the question itself, 'anger' and 'sadness' are both negative sentiments, but clearly, they are different emotions. Emotion analysis can thus be done as an additional layer on top of the (relatively) simple sentiment classification.

===============================================================================
| Sentiment Polarity | 10 Emotion Classes | Emotion word examples             |
|-----------------------------------------------------------------------------|
|                    |    Happiness       | happy, joyful, glad               |
|      Positive      |    Pleasantness    | pleasant, enjoy, nice             |
|                    |    Relief          | relief, comfort, solace           |
|-----------------------------------------------------------------------------|
|                    |    Fear            | scare, fear, frightening          |
|                    |    Sadness         | sad, grief, mourn                 |
|                    |    Disappointment  | bummer, regret, dejected          |
|      Negative      |    Unpleasantness  | dislike, yuck, irksome            |
|                    |    Loneliness      | lonely, withdrawn                 |
|                    |    Anxiety         | anxiety, worry, distress          |
|                    |    Anger           | anger, furious, antagonize        |
===============================================================================

Note that while sentiment classification can be binary, ternary or a graded system (of how +ve or -ve a sentiment is), emotion classification is much more open to interpretation. For example, one could add love as a positive emotion and hatred as a negative emotion. Keep in mind, however, that as we increase the number of classes, building an accurate classifier becomes more and more difficult. You may have to then start thinking of soft clustering approaches because the boundary between two emotions may not be clear (not just AI, it's sometimes not clear even to humans!).

Chthonic Project
  • 8,216
  • 1
  • 43
  • 92
0

It is also called: predicting somebody's mood swings.

I think that Freud would say that emotions are linked to the libido flow (body and soul), while sentiment also has a cultural and moral dimension. Sentiment is already a sublimation of emotion, it is the expression of emotion like in the table below in answer 1. We FEEL glad, happy, scared, angry and express it with word or behaviour patterns.

Negative libido flows cause anger/sadness or joy/happiness. +ve and -ve change happens when external events influence our psychological balance and cause either enhanced happiness or decreased happiness. If know these events and can link them to the expressions of sentiment, you can measure the degree of the change in emotion. The latter depends on the importance of the event in relation with the person's personal set-up (culture, education, psychological set-up).

In a data mining context and looking for intent: if you want to find or even set up a measure for intent you need to find the events that cause patterns in the expressions and usage behaviour of a person. It is like looking for clusters of expressions associated with the person who has the emotions and any specific time and link them to external events. Like many people feel sad or angry about terrorist attacks.

Also, people may actually express sentiments that are contradictory to their emotions. This can only be evaluated if you know cultural/psychological set-up.