I have analysed a number of online articles and have their emotion analysis scores along with their sentiment (pos/neg/neutral) and the sentiment value. The fields are: Anger, Disgust, Fear, Sadness and Joy. What I would like to know if it is possible to somehow combine the values of the fields to represent them as one value. I also have comments related to those articles and have their sentiment and emotion scores as well in similar fields.
This would permit me to find a threshold so that I can use it to grade the article and the comments according to that single value. For example, an article might be: Anger=0.100637, Disgust=0.327951, Fear=0.243857, Joy=0.043951 and Sadness=0.364933.
Clearly in this example, the sadness value is the highest followed by disgust, but would it be right to ignore the lower score fields and classify that article as "sadness" related when "disgust" is that close? Would the "sadness" value be representative of that article? And what if another has 0.148988, 0.14043, 0.070271, 0.609123 and 0.103031? Equal parts "anger" and "disgust" but with 60% "Joy"?
My first thought was to have some sort of mean but that would not be accurate at all as the difference in the different scores will certainly be lost.
Can someone please help me a little with this problem? Can all five values somehow be represented as one? Thank you.
On your first example; when forwarding your interpretation, you could say that the article was mostly associated with sadness and disgust, nothing wrong with that. You could also add that the article was least associated wtih joy and anger.
On the other hand, you may decide to use one emotion (presumably the one with the highest value) in your interpretation of the article only.
A third option would include to simply combine both methods by setting a threshold. By that, I mean setting a value, that will determine when you use one or an n-number of emotions in your interpretation. Let's say in your example you use a threshold of 0.05. Since the difference between disgust and sadness is below that threshold, you include both emotions in your interpretation..."The articles was mostly associated with disgust and sadness." And since the other values are above, they no longer need to be included in your interpretation.
Make sure to state which methodology you will be using for your interpretations. The point is to use that same method for all of your evaluations, once stated. As such, there should be no problem, regardless of which option you choose.
@Mitja - Thank you for your interesting reply. My evaluations or classification will mostly be on numerical data. As you interestingly suggested, I might use a threshold where if two emotion categories (or more) fall within the threshold then I can use both of them to find a single value with which to represent that article.
I could, for example, use PCA feature reduction to combine both sadness and disgust together if they fall within the set threshold. But this brings another question to my mind: under which category would that article be classified? What I'm referring to is a sort of binary value similar to sentiment - positive/negative/neutral. My final interest is to classify articles so that I can compare them to any comments they might have in reply so that an offensive comment is found.
Similarly, if there is only one category that by far has the largest value, then that article belongs to that category.
Hi Silvio, another option would be to use the thresholds, as suggested by Mitja, but as exclusion criteria for your selection of articles. That way you could exclude any articles that were too mixed in the range of emotions expressed.
Regards,
Christian
- Badges
- Science topic
- Similar topics
- Psychology